This paper is a product of the Poverty and Equity Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and contribute to development policy discussions around the world. The authors may be contacted at xluo@worldbank.org, yw548@cornell.edu, and x.zhang@nsd.pku.edu.cn. The Poverty & Equity Global Practice Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. ‒ Poverty & Equity Global Practice Knowledge Management & Learning Team This paper is co-published with the World Bank Policy Research Working Papers. E-Commerce Development and Household Consumption Growth in China Xubei Luo, Yue Wang, Xiaobo Zhang JEL Codes: L81, E21, D23, R12 Key Words: E-commerce, consumption growth, transaction costs, spatial inequality   Xubei Luo is a senior economist of the World Bank; Yue Wang is a Ph.D. candidate at the Cornell University; and Xiaobo Zhang is a distinguished chair professor of economics at the National School of Development, Peking University in China, and senior research fellow of International Food Policy Research Institute. The authors thank Shaohua Chen, Samuel Freije-Rodriguez, John Giles, Bert Hofman, Deepak Mishra, Ambar Narayan, Hoon Sahib Soh, and Salman Zaidi for valuable comments and discussions; Zhengwei Jiang, Zhenzhong Sheng, Ruidong Zhang, and researchers from AliResearch for suggestions and for sharing the Alibaba online data for the research; and Xuejiao Xu for research support. The views expressed here are those of the authors and do not reflect those of the World Bank, its Executive Directors, or the countries they represent.  I Introduction The first big move in online sales came with the founding of Amazon.com in 1994. At the time, the company sold only one product, books, but founder Jeff Bezos had bigger plans to sell “Anything with a capital A.” By 2016, the company had succeeded in that plan and sold $136 billion in goods.1 Meanwhile, in 1999, not long after Bezos launched Amazon, Jack Ma established Alibaba.com, a business-to-business marketplace based in China. In 2018, Alibaba set a milestone for online sales with its Singles’ Day sale, which netted $30.8 billion in transactions, doubling the combined sales for the two biggest sales days in the United States, Black Friday and Cyber Monday.2 E-commerce is a new technology and business mode that allows buyers to make online transactions and receive local package delivery or pickup from sellers. It has reshaped consumption patterns in the years since it was introduced. The welfare implications of e- commerce have been discussed in the literature, mostly focusing on developed countries (Brown and Goolsbee 2000, Hortaçsu et al. 2009, Gorodnichenko and Talavera 2017). On the one hand, e-commerce offers consumers lower search cost and more product variety than traditional retail stores, improving their welfare. Consumers may also gain from accessing merchants online who do not have local brick-and-mortar stores (Dolfen et al. 2019). On the other hand, the intense competition of e-commerce can result in efficiency and welfare gains for the entire society through effects on prices. However, the intense competition may also squeeze out brick-and- mortar stores and smaller retailers, which in turn depresses offline shopping. For example, facing online competition, many physical bookstores have closed (Goldmanis et al. 2010). However, there is a net market expansion effect for consumer electronics, although e-commerce diverts sales from offline, and consumers benefit more than firms from the introduction of e-commerce (Duch-Brown et al. 2017). The net impact of e-commerce depends upon the relative importance of the above factors. In developing countries, a large proportion of people live in remote areas with limited access to offline retail stores. In these countries, e-commerce may have greater potential to reach a wider range of consumers who are otherwise constrained by limited access to markets than in developed countries. Building a multiregional general equilibrium model and using city-level data, Fan et al. (2016) show that e-commerce development in China disproportionately improves consumer welfare in remote cities. Drawing data from eight counties in three provinces where                                                              1 http://www.latimes.com/business/la-fi-amazon-history-20170618-htmlstory.html# 2 https://www.businessinsider.com/black-friday-cyber-monday-vs-singles-day-sales-2018-12.   ··2    Alibaba’s Rural Taobao program was present, Couture et al. (2017) finds that e-commerce expansion reduces the cost of living for certain groups of the rural population who are induced to use it, though the average effect is muted. Following the same spirit of Fan et al. (2016) and Couture et al. (2017), our paper examines the relationship between e-commerce development and consumption in China. We extend the model of Startz (2018) to investigate the differential impact of e-commerce on consumption growth by various goods and services. Compared with Fan et al. (2016) and Couture et al. (2017), our paper has a few new features. First, by matching a nationally representative China Family Panel Studies (CFPS) survey with county-level e-commerce information obtained from Alibaba, we can directly examine the impact of e-commerce development on consumption growth at the household level rather than at the aggregate city level as in Fan et al. (2016). Second, our findings are likely more representative because CFPS covers many more counties than the sample used in Couture et al. (2017). Third, using the rich consumption information of the CFPS survey, we can study the heterogenous associations between e-commerce development and various categories of consumption, which were not discussed in Fan et al. (2016) and Couture et al. (2017). Our paper offers three major findings. First, e-commerce development is associated with higher consumption growth. Lower search cost is a key feature of e-commerce (Bajari and Hortaçsu 2003, Hong and Shum 2006, Brynjolfsson, Dick and Smith 2010, Lieber and Syverson 2011, Levin 2011). Lower search cost makes price discovery easier, bringing the law of one price closer to reality (Gorodnichenko and Talavera 2017). Lower transaction costs increase the level of specialization in the society and create more trade. More competitive prices tend to reduce the cost of living for residents. Couture et al. (2017) show that the expansion of e-commerce to the Chinese countryside is associated with lower cost of living, and for the goods that are available at both the Rural Taobao online terminal and in the village, the median price from the online terminal is cheaper by 15 percent. According to a McKinsey report (McKinsey, 2013), e-tailing may have lowered China’s average retail price by 0.2 to 0.4 percent in 2011 and 0.3 to 0.6 percent in 2012. Holding disposable income constant, lower cost of living means more discretionary spending power, which implies higher consumption. Second, the impact on consumption is conspicuously larger for the rural residents, inland regions, and the poor households, suggesting that e-commerce development helps reduce spatial inequality in consumption. For people in remote areas with limited access to markets, the saving in search costs and the increase in variety of products accessible online compared with in ··3    traditional brick-and-mortar stores can be particularly large. E-tailing is not just a replacement of purchases that would otherwise take place but could spur incremental consumption particularly in small cities and towns where there is pent-up demand for goods that local physical stores cannot deliver (McKinsey, 2013). Couture et al. (2017) show that 62 percent of goods bought through Alibaba’s Rural Taobao online terminal were not available in the village, which rises to 84 percent for durable goods. Therefore, they likely benefit more from e-commerce development than their counterparts in more populous and developed regions. Third, the consumption of durable goods and in-style goods exhibits stronger growth than the consumption of local services. Startz (2018) examines the search and contracting frictions in trade and their role in goods for which sourcing costs and style evolution are of different importance in traditional international trade. Under this framework, by enabling traders to locate nearby producers and sell products directly to consumers, e-commerce can lower travel costs and contracting costs associated with remote ordering. Thanks to the lower costs of travel and contracting, sellers can source more frequently and provide more a la mode products and greater variety to consumers. If the cost saving is passed through, consumers will also get lower prices. As the importance of stylishness and degree of cost savings from search and contracting frictions vary by goods and services, the impact of e-commerce on consumer welfare is likely to differ by types of goods and services. Our finding is accordance with the prediction of Startz (2018). Following this introduction, the paper consists of the following parts: Section II describes e- commerce development in China; Section III discusses the theoretical motivation; Section IV presents the empirical model and data; Section V discusses the empirical results; Section VI conducts robustness analysis; and Section VII concludes. II E-commerce development in China China has quickly become the largest e-commerce market in the world. The number of Internet users in China reached 772 million in 2017, 3 of which 533 million made purchases online.4 The annual total e-commerce trade volume in China increased thirtyfold from RMB 930                                                              3 China Internet Network Information Center (CNNIC) (2018). 41st Statistical Report on the Internet Development in China. http://www.cac.gov.cn/files/pdf/cnnic/CNNIC41.pdf 4 Ministry of Commerce (2017). E-commerce in China 2017. http://cif.mofcom.gov.cn/cif/html/upload/20181101112235744_2017%E5%B9%B4%E4%B8%AD%E5%9B%BD%E7 %94%B5%E5%AD%90%E5%95%86%E5%8A%A1%E6%8A%A5%E5%91%8A.pdf   ··4    billion in 2004 to RMB 29,160 billion in 2017, a compound annual growth rate of 30 percent.5 Express mail service exceeded 30 billion pieces in 2016,6 of which some 60 percent is related to e-commerce.7 According to a 2017 McKinsey report, China’s worldwide e-commerce transaction value grew from less than 1 percent a decade ago to over 40 percent now, exceeding that of France, Germany, Japan, the United Kingdom, and the United States combined (Woetzel et al. 2017). Online retail sales have grown even faster, from RMB 125.7 billion in 2008 to RMB 5,155.6 billion in 2016 – only 1 percent of total retail sales of consumer goods was purchased online in China in 2008, compared with 16 percent in 2016. The number of packages sent through online sales increased tenfold from one billion in 2006 to 10 billion in 2014 (Goldman Sachs 2018). According to the State Post Office, Chinese express firms delivered 40 billion packages in 2017, the majority related to e-commerce.8 The development of online retailing, however, remained uneven across Chinese provinces. In 2015, in Beijing, 45 percent of the total retail sales of consumer goods was purchased online, followed by nearly 40 percent in Shanghai, 35 percent in Zhejiang, and 28 percent in Guangdong. However, this share was much lower (less than 2 percent) in nine inland provinces (Figure 1), as measured by the Online Business Index (OBI) and the Online Shopping Index (OSI) developed by AliResearch (map 1 and map 2).9                                                              5 Ministry of Commerce of China. E-Commerce in China (2015, 2016, and 2017). 6 Source: National Bureau of Statistics of the People’s Republic of China, http://www.stats.gov.cn/tjsj/sjjd/201706/t20170626_1506952.html 7 Source: State Post Bureau of the People’s Republic of China, http://www.spb.gov.cn/zy/xxgg/201706/t20170624_1196398.html 8http://www.spb.gov.cn/xw/dtxx_15079/201806/t20180604_1581131.html 9 The OBI and OSI data are obtained from the AliResearch team. OBI is a constructed index measuring the density of online stores and the percent of online stores with annual online sales above RMB 240,000. The value of the index ranges from 0 to 100. The higher the value, the more developed the online sales. OSI is a constructed index measuring the density of online buyers and the percent of online buyers with annual online consumption above RMB 10,000. The value of the index ranges from 0 to 100. The higher the value, the more developed the online purchases. The online transaction numbers are from the Alibaba platform, which accounts for the majority of online transactions in China. The values of the OBI and OSI are not comparable over time due to changes in the methodologies. ··5    Map 1. Online Business Index (county level) 2013 2014 2015 Data source: AliResearch Map 2. Online Shopping Index (county level) 2013 2014 2015 Data Source: AliResearch ··6    There is also a gap in the development of online retail between urban and rural areas. Up to December 2016, the rural Internet users in China accounted for 27.4 percent of the national total (731 million), while urban Internet users took up 72.6 percent of the total, much higher than the share of urban population (57 percent). The Internet penetration in urban areas was 69.1 percent, while in rural areas it was only 33.1 percent. Nearly three-quarters of online stores and Internet users were concentrated in urban areas.10 From a positive perspective, the wide gaps across regions and between rural and urban areas imply large growth potential in the less developed areas. In fact, total online retail transactions have grown faster in rural areas than in urban areas in the past several years. Rural online retail transactions increased from RMB 353 billion in 2015 to RMB 895 billion in 2016, representing 17 percent of the total online retail transactions, an increase from 9 percent in 2015.11 Figure 1. Share of online retail sales to total retail sales of consumer goods in provinces (2015) 50% Retail Sales of Consumer Goods (2015) Share of Online Retail Sales to Total  45% 40% 35% 30% 25% 20% 15% 10% 5% 0% Qinghai Shaanxi Hubei Guizhou Xinjiang Heilongjiang Beijing Tianjin Fujian Sichuan Hainan Chongqing Hunan Shandong Henan Liaoning Gansu Jilin Zhejiang Anhui Hebei Jiangxi Guangxi Ningxia Inner Mongolia Tibet Shanghai Guangdong Jiangsu Yunnan Shanxi Source: Staff calculations based on China Statistical Yearbook 2016 III Theoretical motivation: Extension of Starts (2018) In the model developed in Startz (2018), consumers care about both quantities and style of the goods they consume. Traders choose whether to order remotely from suppliers or travel to the suppliers. They also need to decide the optimal stocking period (frequency) to maximize average profit over a period T, taking travel cost and the cost of search and contracting associated                                                              10 Ministry of Commerce of the People’s Republic of China 2016. “E-commerce in China, 2016” (中国电子商务报告). 11 Calculated based on data from the National Bureau of Statistics of China and the Ministry of Commerce of the People’s Republic of China, “E-Commerce in China.” Total rural online retail transactions refers to the sum of online retail transactions from e-commerce enterprises (including individuals) operated in the administrative regions at the county level or below (excluding city districts) (http://images.mofcom.gov.cn/dzsws/201706/2017061110205702.pdf). ··7    with remote ordering into account. The optimal stocking period is shorter for varieties evolving more frequently and in greater demand. This holds true also for traders with lower travel cost as the style frontier can only be observed when traders travel to the production location. In a world without e-commerce, traders will choose to travel to the production site if the loss due to search friction and contracting premium overtakes the travel cost, and order less up-to-date varieties remotely otherwise. E-commerce provides a third option for traders, that is, to locate in the same location as suppliers and sell remotely online to consumers. In this way, they can observe the frontier style without the travel costs, facing lower search friction and contracting costs. In addition, there is a saving in fixed storage cost as they can directly source from suppliers to meet the online order. Therefore, they can keep a limited inventory with minimum cost. In summary, e-commerce can improve consumer welfare in three ways. First, they provide more up-to-date goods. Second, in the face of lower entry cost, more traders will enter the market and sell more varieties. Third, the intense competition may force traders to pass part of the cost saving to consumers, leading to lower prices, which directly improve consumer welfare. Under this framework, associated with e-commerce development, consumers will purchase more goods that evolve faster in style, such as cosmetics and clothes, as well as durable goods that have high storage costs, particularly in remote areas with higher travel costs. Moreover, e-commerce may yield an indirect income effect thanks to the lower price. Lower prices for goods purchased online mean that consumers have more disposable income. The extra income is likely to be spent on goods and services with high income elasticity not available online, such as travel and dining out. However, some personal services, such as kindergarten and health care, require frequent interactions between suppliers and consumers. In this case, local service providers enjoy a greater advantage than remote ones. As a result, e-commerce development likely will have little effect on such locally provided personal services.12 IV Empirical method and data description                                                              12 Our discussion here refers to personal services where the product cannot be separated from the supplier, such as teaching, medical service, and personal care. Services such as typing, photo processing, and consulting are not included. The latter kinds of services, where the product can be separated from the labor provider, share the same features as goods. ··8    In this paper, we examine the relationship between e-commerce development and household consumption growth in China using three data sources: Alibaba e-commerce data, household survey data, and official statistics. First, we construct a series of indicators to measure the development of e-commerce at the county level using online sales and purchase information provided by the Alibaba Group, gross domestic product (GDP) from the China Statistical Yearbook, and population drawn from the Population Census 2010. Second, we merge the county-level e-commerce development measures with household consumption data obtained from the CFPS survey administered by Peking University. Indicator construction We measure county-level e-commerce development on three dimensions: penetration, intensity, and market size (as a share of the national online market). For each of the three dimensions, we construct two indicators measuring, respectively, the online sales and purchase using the amount of online sales and online purchases (Gross Merchandise Value, or GMV), numbers of online sellers and online purchasers, and numbers of packages sent and received associated with online transactions. The three indicators are comparable over time. Therefore, compared with Alibaba’s three official e-commerce indicators, Alibaba e-commerce development indicators (aEDI), OBI, and OSI, which use different weights across years, our three indicators are comparable over years. We constructed the indicators in three steps. First, we extracted the original indicators for each index. Second, we normalized the indicators and transformed them into z-scores for easy comparison.13 The number of buyers, annual online purchase GMV, national online purchase GMV, and total number of national online buyers are from Alibaba. GDP is obtained from China Statistical Yearbooks. Residential population is from the Population Census 2010.14 The three sets of indicators are defined as follows: E-commerce penetration indexes: Share of online buyers in population (%): 100 Share of online purchases in GDP (%): 100                                                              13See Appendix 1 for a detailed description of the construction of the indicators.  14Measures using different sources of population yield consistent results. We use census population information in our main analysis, rather than the population reported in the statistics year books. In addition, we apply the 2013 residential population at the county level collected from local statistical bureaus as a robust check. ··9    E-commerce intensity indexes: Per buyer online purchases (yuan): Per capita online purchases (yuan): Market size indexes: National share of online purchases (%): 100 National share of online buyers (%): 100 Each of the measures is normalized into a z-score: z score of measure x . 15 Similar to the OBI and OSI, online purchase and sales intensity (defined as the online purchase GMV or online sales GMV over census population) varies widely across regions.16 The z-score mean of online purchase intensity in CFPS counties over the period 2013 to 2016 is 0.39 standard deviation above the mean in the east, 0.21 standard deviation below the mean in the central region, and 0.44 standard deviation below the mean in the west. At the same time, the z- score mean of online sales intensity over 2013 to 2016 is 0.38 standard deviation above the mean in the east, 0.27 standard deviation below the mean in the central region, and 0.36 standard deviation below the mean in the west. Analytical models We examine the relationship between e-commerce and household consumption growth, controlling for the effects of household consumption of the initial year, key household characteristics (including age and dependency ratio), and regional characteristics (including regional dummies of east, central, and west, as well as urban and rural). Our household data are                                                              15 The indicators are strongly correlated with one another, we will present the first indicator, e-commerce intensity measured by online purchases per capita (online purchases over census population) in the main model and use the others as robust checks (see discussion in the following section). 16 We use the population information from the census.    ··10    from CFPS 2014, and 2016, covering 156 counties in China.17 The CFPS data provide household characteristics, total household consumption, and detailed information on 18 categories of subcomponents.18 Our basic model is specified as below: Y , , , , , , /10 , , , where i is the index for household, c is county, t is year, s is consumption category. The dependent variable Y , , , is the growth of log consumption per capita of category s in household i located in county c between two waves of the CFPS survey. It is the difference between two adjacent waves –2014 and 2016. Our variable of interest is , . It is the z-score of the e-commerce development indicators. To measure the consumer side of e-commerce development in county c, we start by using , , online purchases over the 2010 census population in 19 county c in 2013. For readability, we present the empirical results of the online purchases intensity indicator, , in the main analysis, and the results of other e- commerce development indicators – another measure of the intensity indicators, two other measures of the penetration indicators, and two other measures of the market size indicators – as robustness checks.20 We include, besides the dummy variables to capture the difference between the east, central, and west regions, and between rural and urban areas, the following variables of control at the household level: , : the lag of log consumption per capita of category s in household i located in county c, that is the consumption per capita two years ago obtained from the previous wave of CFPS.                                                              17 The CFPS covers 162 counties in 25 provinces. Due to administrative change, eventually, we are able to merge 156 counties with our county-level e-commerce data from Alibaba. 18 The 18 categories of consumption do not add up to the total household consumption, but they comprise the majority of consumption. 19 The e-commerce data in 2013 are the earliest year available from Alibaba. We use this to minimize reverse causality between e-commerce development and household consumption growth.  20 Results are largely consistent with those of the main indicators. Details available upon request. ··11    , : the dependent ratio in household i located in county c in year t. , , : the average adult age in household i in county c, year t. , : 1 if household i is an urban household and 0 otherwise. : the region fixed effect according to the National Bureau of Statistics region definition of east, west, and central China. : the error term. Standard errors are clustered on the county level to account for correlation within a county. Our extended model includes more control variables to capture the possible impact on household consumption growth of other key factors, with cross-section data for 2014-2016: Y , , , , , , /10 /100 , ℎ , , , The additional variables of control include: ℎ , : average years of schooling in household i, located in county c, year t. , :is the male over female gender ratio in household i. : the year fixed effect. 21 We run the model for per capita total consumption, different consumption components, and subsamples by quartile, respectively. Drawing from the information in CFPS, we include 18 categories of household consumption: cosmetics and beauty, food (which includes two subcategories, food at home and dining out), clothes, utilities, communications, local transport, travel, entertainment, automobiles, other vehicles, durable goods, medical, health and fitness, education, home repairs, and gifts.                                                              21 We include year fixed effects for total consumption and most of the consumption categories, except house repair expenditure and gift expenditure. Since there are no questions about house repair expenditure and gift expenditure in CFPS 2012, we cannot calculate the growth of these two categories for 2012–2014. Therefore, for these two consumption categories, we do not include year fixed effects in the model.   ··12    V Empirical Results The growth rate of household consumption per capita is higher when online purchasing intensity measured by online purchases over population is greater, others being equal (Figure 2). Figure 2. Growth rate of household consumption per capita and e-commerce intensity level in China Note: Extreme value dropped at 5 percent. Table 1 presents the results from our basic model nationwide and by rural-urban areas, as well as by regions. The role of e-commerce development is significant in all estimations. A one standard deviation above the mean of e-commerce intensity measured by purchases over census population is associated with 0.303 increase in log annual household expenditure per capita growth. The associated increase is more than twice larger among rural households (0.718) than that among urban households (0.247). Across regions, the magnitude is the largest in the west (0.538), followed by the central region (0.313) and the smallest in the east (0.267). The variables of control are of the expected signs. The log of household consumption expenditure in the initial year is negatively associated with household consumption growth in a significant manner. Average age of household members and its square terms are both positively associated with household consumption growth, while the dependency ratio is negatively associated with household consumption growth. ··13    Table 1. E-commerce development and total household expenditure per capita (1) (2) (3) (4) (5) (6) VARIABLES All Urban Rural East Central West Z-score of purchase/ 0.303*** 0.247*** 0.718*** 0.267*** 0.313** 0.538*** population (0.067) (0.061) (0.098) (0.083) (0.119) (0.157) Lag log household -0.596*** -0.548*** -0.664*** -0.581*** -0.582*** -0.641*** expenditure per capita (0.015) (0.019) (0.016) (0.026) (0.024) (0.025) Average age of household -0.057*** -0.049*** -0.099*** -0.064*** -0.056*** -0.068*** members/10 (0.009) (0.014) (0.014) (0.015) (0.017) (0.019) Square of average age of 0.001*** 0.000*** 0.002*** 0.001*** 0.001** 0.001*** household members/100 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Household dependent ratio -0.131*** -0.045 -0.244*** -0.103** -0.133*** -0.185** (0.033) (0.039) (0.044) (0.049) (0.043) (0.090) Urban/rural area dummy 0.242*** 0.254*** 0.204*** 0.302*** (Urban=1) (0.028) (0.055) (0.043) (0.050) West 0.033 0.047 0.035 (0.051) (0.061) (0.059) Central 0.063 0.035 0.111** (0.046) (0.055) (0.048) Constant 6.092*** 5.795*** 7.152*** 5.963*** 6.038*** 6.720*** (0.169) (0.227) (0.187) (0.258) (0.292) (0.297) Observations 11,107 5,183 5,924 4,457 3,486 3,163 Adjusted R-squared 0.319 0.285 0.368 0.319 0.321 0.285 0.3010.368 0.321 0.338 0.301 0.33 Note: Robust standard errors clustered at the county level in parentheses. *** p<0.01, ** p<0.05, * p<0.1 We examine the pattern of association between e-commerce development and each of the 18 categories of subcomponents of household consumption in Table 2. For most of the 18 categories of consumption, the association with e-commerce development is positive, stronger in the less developed regions and in rural areas. Three patterns are apparent from the table. Table 2. E-commerce development and various categories of household expenditure per capita by region (1) (2) (3) (4) (5) (6) All Urban Rural East Central West Cosmetics and beauty 0.490*** 0.449*** 0.983*** 0.510*** 0.415*** 1.391*** (0.121) (0.125) (0.235) (0.162) (0.152) (0.476) Food 0.355*** 0.288*** 0.842*** 0.336*** 0.352*** 0.673** (0.066) (0.061) (0.118) (0.081) (0.111) (0.298) Of which: Food at home 0.389*** 0.348*** 0.805*** 0.357*** 0.412*** 0.606** (0.073) (0.066) (0.144) (0.094) (0.111) (0.263) ··14    Of which: Dining out 0.580*** 0.501*** 1.260*** 0.423** 0.777*** 1.528*** (0.164) (0.165) (0.267) (0.211) (0.231) (0.495) Clothes 0.306*** 0.269*** 0.720*** 0.257* 0.376*** 0.590** (0.102) (0.103) (0.163) (0.131) (0.125) (0.287) Utilities 0.181*** 0.180*** 0.275** 0.183*** 0.154*** 0.469 (0.033) (0.032) (0.138) (0.040) (0.047) (0.314) Communications 0.242*** 0.210*** 0.565*** 0.184*** 0.309*** 0.545*** (0.051) (0.050) (0.078) (0.058) (0.069) (0.198) Local transport 0.435*** 0.384*** 0.732*** 0.328*** 0.502*** 0.518** (0.092) (0.093) (0.138) (0.115) (0.096) (0.253) Travel 0.872*** 0.770*** 1.595*** 0.751** 0.899*** 2.131*** (0.211) (0.214) (0.250) (0.283) (0.157) (0.583) Entertainment 0.655*** 0.547*** 1.171*** 0.546** 0.852*** 0.441 (0.169) (0.174) (0.325) (0.235) (0.163) (0.404) Automobiles 0.474*** 0.394*** 1.406*** 0.285* 0.615*** 1.329*** (0.127) (0.115) (0.291) (0.148) (0.202) (0.323) Other vehicles 0.182 0.116 0.841** 0.071 0.380 0.162 (0.136) (0.126) (0.377) (0.137) (0.371) (0.566) Durable goods 0.444*** 0.357*** 1.029** 0.371** 0.420** 0.621 (0.118) (0.104) (0.397) (0.158) (0.181) (0.958) Medical 0.166** 0.154* 0.309 0.071 0.287** 0.672 (0.075) (0.085) (0.201) (0.086) (0.127) (0.432) Health and fitness 0.395*** 0.327*** 0.534*** 0.351* 0.381*** 0.636** (0.126) (0.123) (0.195) (0.188) (0.138) (0.299) Education -0.102 -0.091 0.008 -0.157* 0.100 -0.165 (0.118) (0.125) (0.274) (0.091) (0.204) (0.402) Home repair -0.043 -0.072 0.116 -0.001 -0.109 -0.280 (0.083) (0.086) (0.314) (0.131) (0.149) (0.707) Gifts 0.061 0.073 0.454 0.013 0.210* 0.281 (0.107) (0.099) (0.276) (0.140) (0.120) (0.545) Note: The model of estimation is the same as Table 1, the dependent variable is the logarithmic form of household per capita consumption growth, and the variables of control include lagged log value of household expenditure per capita, average age/10, age squared/100, dependency ratio, urban-rural dummies, and regional dummies. For readability, we present only the results of interest - the coefficients of the z-score of online purchase / population in 2013. Fully results available upon request. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. First, for goods and services regularly consumed by most households,22 including expenditures on cosmetics and beauty, food, clothes, utilities, communication, local transportation, and entertainment, we see a positive association between the z-score of online purchases over                                                              22 From the CFPS data, over 50 percent of the households reported such consumption expenditures. ··15    population and log expenditure growth across both rural and urban areas and all three regions in most cases.23 The coefficients are larger among rural households and in the interior region than in urban areas and in the east systematically.24 A one standard deviation increase of online purchases over population is associated with a 0.490 increase in growth of log annual household expenditure per capita on cosmetics and beauty products.25 The magnitude is larger among rural households (0.983) than that among urban households (0.449). And the magnitude is the largest in the west, which is as high as 1.391. The relationship between the z-score and food consumption follows a similar pattern. When online purchases over population increases by one standard deviation above the mean, log household food consumption per capita growth increases by 0.355. The increase is 0.288 among urban households and almost three times that among rural households (0.842). In the east, it is 0.336 while in the central, it is similar to that in the east (0.352) and in the west, it is twice as large as that in the east (0.673). When purchases over population increases by one standard deviation above the mean, log household dining out consumption per capita growth increases by 0.580. In rural areas, it is as high as 1.260 and, in the west, it is as high as 1.528. A one standard deviation increase of online purchases over population is associated with 0.306 increase in growth of log annual household expenditure on clothes. The magnitude among rural households is almost threefold as large as that among urban households (0.729 and 0.269 respectively for rural and urban). And across the three regions, it is the largest in the west, which is 0.590, followed by the central region, which is 0.376 and is the smallest in the east, which is 0.257. When the z-score increases by one, log household communication consumption per capita growth increases by 0.242 for the whole sample, 0.210 for urban households, 0.565 for rural households, 0.184 in the east, 0.309 in the central region, and 0.545 in the west.26 When the z-score increases by one, log household entertainment consumption per capita growth increases by 0.655 for the whole sample, 0.547 for urban households, 1.171 for rural households, 0.546 in the east, 0.852 in the central region, and 0.441 in the west. Second, for goods and services that require a larger payment at one time and are less frequently purchased,27 i.e., travel, automobiles, and durable goods, we also see a positive                                                              23 The coefficient of health and fitness expenditure is about 0.3 for all households. As only roughly 10 percent of households reported health and fitness expenditure, we do not discuss this in the paper. 24 The coefficient of utility expenditure is about 0.18 for all households, smaller than those of other variables. This might be related to the nature of utility consumption. 25 Cosmetic and beauty products, including a wide range in prices from the very low end in the spectrum, are becoming a popular purchase online, in particular, among young female online buyers. 26 We do not have information on whether dining out refers to sit-down dining only or includes fast food consumed outside or home delivery. One possible explanation for the large coefficient of dining out is the rapid increase of the availability of online food delivery. 27 From the CFPS data, less than 50 percent of the households reported such consumption expenditures.  ··16    association between the z-score of online purchases over population and log expenditure growth across both rural and urban areas and all three regions in most cases. The magnitude of the coefficient is larger, above 0.4 for all households, and the coefficients of some categories can be as high as 2 or more for rural households and households in the west. The correlation between log automobile purchases consumption growth and the z-score of online purchases over population is 0.474 for the whole sample, 0.394 among urban households, 1.406 among rural households, 0.285 in the east, 0.615 in the central region, and 1.329 in the west. Similarly, a one standard deviation increase above the mean of online purchases over population is associated with 0.444 of the log durable goods consumption growth all over the country. The coefficient is 0.357 among urban households and 1.029 among rural households. Across regions, the coefficient is 0.371 in the east, 0.420 in the central region, but not statistically significant in the west. Third, for personal services, such as expenditure on vehicles other than purchasing automobiles,28 household repair, and education, the role of e-commerce intensity, measured by online purchases over census population, is not significant.29 However, for health and fitness, there is a positive association between the z-score of online purchases over population and across urban and rural, as well as the three regions, since health and fitness consumption could be a combination of both personal service and purchased goods such as equipment and supplements. The magnitude is also larger in rural areas in the west. As for medical expenses, which is also a combination of personal services and purchased goods, with the former more important, we see a positive association among the population, urban households and in the central region, but not among rural households, neither in the east nor the west. Table 3 presents the results of the basic model at the national level by household expenditure levels.30 For the growth in total household consumption per capita, e-commerce matters the most for the poorest quartile with a coefficient of 0.266, followed by the 25 percent– 75 percent quartile with a coefficient of 0.198, compared to a coefficient of 0.126 for the richest quartile. E-commerce purchase intensity has a positive effect in 12 of 16 household consumption categories and the effect is the strongest among the poorest quartile for half of them (cosmetics and beauty, food, utilities, travel, medical, health and fitness). For three household consumption                                                              28 Consumption on vehicles other than automobiles might also fall into this category, while the coefficient is smaller than 0.5. It is a combination of buying goods and personal services, as it includes repairing vehicles and communication devices. As for consumption on vehicles other than automobile purchase, the coefficient is not significantly different from zero for the whole sample, across the three regions and among urban households, 0.841 among rural households, as shown in Table 2. 29 We do not have information on whether the slower growth in education expenditure is related to the availability of free or lower priced online information and books, or other emerging patterns on education expenditure. 30 Extreme value is dropped at 1 percent.  ··17    categories, including cosmetics and beauty, food, , and travel, e-commerce purchase intensity is associated with positive consumption growth for households of all income groups, and for two household consumption categories – medical, and health and fitness consumption – the effect is positive only for the poorest quartile. The effect of e-commerce on the consumption of durable goods and vehicles other than automobiles is not significant in any income group. For two household consumption categories – communication and entertainment , the magnitude of the correlation is the largest for the middle-income group. For local transport and automobile purchases, it is the largest for the top income group. Similar to the results of Table 2, the effect of online purchases over population is not significant for most consumption groups in three categories: education, home repair, and gifts. Table 3. E-commerce development and various categories of household expenditure per capita by household consumption level (1) (2) (3) (4) All Bottom 25% 25%-75% Top 25% Total household expenditure 0.303*** 0.325*** 0.157*** 0.112*** (0.067) (0.091) (0.049) (0.037) Cosmetics and beauty 0.490*** 0.568** 0.239* 0.179** (0.121) (0.227) (0.133) (0.086) Food 0.355*** 0.489*** 0.253*** 0.129*** (0.066) (0.120) (0.063) (0.042) Of which: Food at home 0.389*** 0.516*** 0.293*** 0.195*** (0.073) (0.148) (0.077) (0.056) Of which: Dining out 0.580*** 0.274 0.202 0.208* (0.164) (0.220) (0.159) (0.125) Clothes 0.306*** 0.199 0.068 0.013 (0.102) (0.223) (0.090) (0.082) Utilities 0.181*** 0.294** 0.128*** 0.041 (0.033) (0.124) (0.036) (0.028) Communications 0.242*** 0.028 0.138** 0.111*** (0.051) (0.116) (0.056) (0.031) Local transport 0.435*** 0.175 0.178** 0.257** (0.092) (0.252) (0.080) (0.121) Travel 0.872*** 0.695** 0.443** 0.535*** (0.211) (0.297) (0.174) (0.181) Entertainment 0.655*** 0.133 0.386* 0.367*** (0.169) (0.144) (0.206) (0.115) Automobiles 0.474*** 0.135 0.171 0.326** (0.127) (0.173) (0.136) (0.145) Other vehicles 0.182 -0.088 0.029 -0.017 ··18    (0.136) (0.273) (0.193) (0.113) Durable goods 0.444*** 0.016 0.061 0.092 (0.118) (0.389) (0.115) (0.121) Medical 0.166** 0.552** -0.028 0.166 (0.075) (0.230) (0.116) (0.154) Health and fitness 0.395*** 0.445** -0.006 0.160 (0.126) (0.210) (0.108) (0.117) Education -0.102 -0.392 -0.105 0.080 (0.118) (0.324) (0.115) (0.224) Home repair -0.043 -0.218 -0.091 -0.372*** (0.083) (0.281) (0.120) (0.122) Gifts 0.061 0.501 -0.002 -0.272* (0.107) (0.411) (0.116) (0.144) Note: The model of estimation is the same as Table 1, the dependent variable is the logarithmic form of household per capita consumption growth, and the variables of control include lagged log value of household expenditure per capita, average age/10, age squared/100, dependency ratio, urban-rural dummies, and regional dummies. For readability, we present only the results of interest - the coefficients of the z-score of online purchase / population in 2013. Fully results available upon request. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Similar to the findings in Table 2, the z-score of online purchases over population is associated with an increase of consumption for food, cosmetics and beauty, utilities, clothes, travel, and health and fitness, with the largest increase for the poorest segment of the population. The sizes of the coefficients of the z-scores are moderate. When the z-score of online purchases over population increases by one, log household food consumption per capita growth increases by 0.489 for the poorest quartile, by 0.253 for the middle two quartiles, and by 0.179 for the richest quartile; log household cosmetics and beauty consumption per capita growth increases by 0.568 for the poorest quartile, by 0.239 for the middle two quartiles, and 0.179 for the richest quartile; log household utilities consumption per capita growth increases by 0.294 for the poorest quartile, by 0.128 for the middle two quartiles, but has no significant change for the richest quartile;. When the z-score of online purchases over population increases by one, log household medical expenditure growth increases by 0.552 for the poorest quartile, and log household health and fitness expenditure growth increases by 0.445 for the poorest quartile, while the association among the middle two quartiles and the top quartile for these two categories is not significant and smaller in size. The marginal effects tend to decline as the income quartiles move up. This suggests that online access to a larger selection of products (including potentially products at different quality and price levels) can have a stronger stimulation effect in household ··19    consumption for the poor by unlocking the potential demand for basic goods that are not available (or available at a higher price) at the traditional local markets. When the z-score of online purchases over population increases by one, log household travel expenditure growth is associated with 0.695 increase for the poorest quartile, 0.443 for the middle two quartiles and 0.535 for the richest quartile. For communication and entertainment consumption expenditure, the coefficient for the e-commerce indicator is the highest for the middle two quartiles (0.138 and 0.386 respectively). When online purchases over population increases by one standard deviation for the poorest quartile, log household consumption growth on cosmetics and beauty products increases by 0.568, log household consumption growth on travel increases by 0.695.31 The large magnitudes of the coefficients suggest that online purchases over population has strong effects on the consumption growth of the goods and services of higher income elasticity / higher in style consumption nature (such as travel and cosmetics and beauty products), particularly for the poorer households, whose demand might have been disproportionately limited by their poor access to markets through traditional markets. Consistent with the prediction of the model in Startz (2018), the results show that e- commerce development is associated with stronger growth in consumption of goods with rapidly evolving styles (such as cosmetics and beauty products, clothes, and entertainment), and goods with higher storage costs (such as durable goods that require more resources for stocking for local traders). We also see higher consumption growth of goods and services with higher income elasticities, such as travel and dining out, even though these are not online products or services. In the meantime, there is some indication that e-commerce development benefits consumers in areas with high travel costs (such as in the interior regions). Across the different categories of consumption of goods and services, the association between e-commerce development and food consumption growth is particularly strong. From 2012 to 2016, China's e-grocery sales saw a compound annual growth rate of 52.9 percent, according to a recent report by Agriculture and Agri-Food Canada.32 The extremely strong growth of online food consumption in China seems to defy the Engel curve. The puzzle is likely                                                              31 Health and fitness consumption is different from medical treatment consumption. The former does not include medical necessities but only healthy supplements and fitness while the latter is associated with treatment in hospital or medicines bought from a pharmacy. The results are consistent with expectations as health and fitness expenditure might have a stronger income elasticity while medical expenditure is more need-based. 32 Source: http://www.agr.gc.ca/eng/industry-markets-and-trade/international-agri-food-market- intelligence/asia/market-intelligence/e-grocery-market-in-china/?id=1504037238257.   ··20    due to the large increase in online consumption of baby food. Baby food and formula are the top two package foods sold online in China.33 Using the CFPS data, we find that breast feeding is negatively correlated with e-commerce development; the coefficient between food consumption and e-commerce z-score is larger for households with young children under three years old.34 This is consistent with the Chinese culture and the strong willingness of parents to spend on children, while the judge is still out for the impact on nutrition. The fast growth in food consumption may also stem from the online purchases of food varieties by higher-end consumers. In more developed areas, particularly tier one and tier two cities, people can purchase online specialty fresh produce, meat, or seafood (which are not available in traditional markets) with guaranteed fast delivery. End-to-end value chain digitalization, such as the Alibaba’s Hema store, which sells online high-end groceries and ready- made food with guaranteed fast delivery, is one example that might be changing the pattern of online-offline retails in the future.35 However, the contribution of such high-end online-offline grocery purchases to food consumption is likely to be limited for the period of our analysis given the limited coverage and the high requirement of logistics, although this might increase in the future, particularly in more developed areas. As to personal services, such as education, medical, fitness, and health services for which personal interaction is more important, consumption growth is not associated with e-commerce development.36 These services involve strong personal interactions. For example, high quality of service in education often requires instructor customization of teaching strategy, content, and speed according to students’ needs and feedback. As labor mobility is lower than the mobility of goods, it is costly to send service providers to distant customers. VI Robustness                                                              33 See https://www.newyorker.com/magazine/2018/07/23/how-e-commerce-is-transforming-rural-china 34 Additional results on the relations between food consumption and whether the household has young children and those on the breastfeeding and e-commerce z-scores are available upon request. Forthcoming in another research paper. 35 https://www.pwccn.com/en/retail-and-consumer/publications/global-consumer-insights-survey-2018-china-report.pdf  36  Due to data limitation, we cannot examine the use of online education or health services, such as interactive online learning program for children and health monitoring apps, which may have a spillover effect over time for consumers to consume more of not just online consumption goods but also get exposed to the kind of goods and services that can potentially enhance human capital.  ··21    To check robustness, we conducted a series of alternative specifications to examine the possible effects of other variables on the growth rate of household per capita total consumption as well as that of each of the 18 categories of consumption. We first tested the various indicators measuring the penetration, intensity, and market size of e-commerce from both the purchase and selling aspects. Table A1 in the Appendix presents the empirical results of the baseline model using other constructed indicators measuring e-commerce penetration, intensity, and market size. The results are similar to those presented in Table 1 – e-commerce development, measured in different ways, plays a positive role in shaping household consumption, with a stronger correlation in inland regions and rural areas, and in households with lower income levels. We introduced several additional variables of control to test whether the relation between e-commerce development and household consumption growth is a result of variables missing in the basic model. Tables A.2 and A.3 present the results of an extended model, with additional household control variables, including lagged log value of household expenditure per capita of the consumption category in question, household average years of schooling, household gender ratio, average age, age squared/100, dependency ratio, urban-rural dummies, and regional dummies. Tables A.4 and A.5 present the results of a further extended model, with additional variables of control, including the lagged log household per capita income growth, lagged log value of household expenditure per capita of the consumption category in question, household average years of schooling, household gender ratio, average age, age squared/100, dependency ratio, urban-rural dummies, and regional dummies. Tables A.6 and A.7 present the results of an alternative extended model, with additional variables of control, including the lagged log household per capita income per capita (2010 value), lagged log value of household expenditure per capita in the consumption category in question, household average years of schooling, household gender ratio, average age, age squared/100, dependency ratio, urban-rural dummies, and regional dummies. Tables A.8 and A.9 present the results of another alternative extended model, with additional variables of control, including consumer price index (CPI) by category, lagged log value of household expenditure per capita in the consumption category in question, household ··22    average years of schooling, household gender ratio, average age, age squared/100, dependency ratio, urban-rural dummies, and regional dummies. The results of the alternative models in the robustness tests are largely consistent with those in Tables 1 and 2, which suggests that after controlling for household demographic characteristics, as well as income growth, initial income level, or inflation, the relation between e- commerce development and household consumption growth remains largely the same: e- commerce development is associated with strong household consumption growth. VII Conclusion Our results show that, in China, household consumption growth is positively associated with initial local e-commerce development. The relationship is stronger for households in the less developed inland regions and in rural areas than in the coastal regions and urban areas, in particular for the poorer households than for the richer ones. E-commerce development reduces consumption inequality across regions and income groups. Thanks to the lower search and transaction costs, more products become tradable online at lower price, if the cost saving is passed on to consumers due to competition. The potential decline in price, increase in variety and smaller travel cost are likely the main channels that contribute to an increase in consumer welfare, as argued by Couture et. al. (2018). Income increase associated with e-commerce business development in an area is also a potential channel of the consumption growth. We tested this in Table 4. For the poorest quartile, the coefficient of the z-core of online purchases over population on log household income per capita growth is not significantly different from zero. This implies that the impact of e-commerce on consumption mainly stems from price and variety channels for the poor. We cannot rule out the income channel for other groups. Table 4. E-commerce development and log household income per capita growth (1) (2) (3) (4) (5) (6) (7) (8) (9) VARIABLES All Urban Rural East Central West bottom 25% 25%-75% top 25% Z-score of purchase/ 0.353*** 0.288*** 0.824*** 0.344*** 0.315*** 0.710*** -0.146 0.147*** 0.081** population (0.0701) (0.062) (0.099) (0.087) (0.104) (0.144) (0.128) (0.025) (0.035) Lag log household -0.735*** -0.722*** -0.764*** -0.706*** -0.740*** -0.780*** -0.931*** -0.919*** -0.952*** expenditure per capita (0.0172) (0.024) (0.019) (0.033) (0.023) (0.033) (0.018) (0.008) (0.021) ··23    Average age of -0.006 0.057*** -0.090*** -0.022 -0.003 -0.204 0.000 0.002 -0.003 household members/10 (0.013) (0.018) (0.021) (0.017) (0.025) (0.126) (0.022) (0.007) (0.013) Square of average age 0.000 -0.001*** 0.001*** 0.000 -0.000 0.021 0.000 -0.000 0.000 of household (0.000) (0.000) (0.000) (0.000) (0.000) (0.014) (0.000) (0.000) (0.000) members/100 Household dependent -0.269*** -0.250*** -0.381*** -0.209*** -0.285*** -0.312*** -0.226*** -0.067*** -0.126*** ratio (0.0348) (0.050) (0.050) (0.051) (0.056) (0.077) (0.064) (0.022) (0.039) Average years of 0.059*** 0.068*** 0.047*** 0.064*** 0.050*** 0.067*** 0.014** 0.016*** 0.010*** schooling (0.004) (0.005) (0.005) (0.006) (0.005) (0.009) (0.006) (0.002) (0.004) Urban/rural area 0.185*** 0.158*** 0.190*** 0.216*** -0.020 0.020 0.010 dummy (Urban=1) (0.0312) (0.055) (0.038) (0.069) (0.058) (0.014) (0.031) West -0.081 -0.065 -0.103 -0.002 -0.003 0.015 (0.058) (0.077) (0.062) (0.069) (0.025) (0.047) Central 0.011 -0.011 0.056 0.125* 0.029 -0.021 (0.052) (0.061) (0.058) (0.067) (0.018) (0.038) Constant 6.808*** 6.484*** 7.827*** 6.572*** 6.916*** 7.711*** 7.290*** 8.595*** 10.036*** (0.194) (0.262) (0.245) (0.350) (0.266) (0.288) (0.244) (0.094) (0.221) Observations 10,138 4,709 5,429 4,055 3,224 2,858 2,062 5,180 2,896 Adjusted R-squared 0.501 0.472 0.540 0.469 0.520 0.530 0.756 0.887 0.760 F test 215.7 126 219.5 85.36 168.8 157.4 436 1751 330.9 Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The positive association between e-commerce and consumption growth is significant not only for total household consumption, but also for most consumption categories, ranging from basic consumption, such as food, clothes, and local transportation, to consumption of durable goods, travel, and entertainment. The magnitude of the association of e-commerce intensity is larger for the second consumption group than for the first. For most of the consumption categories, the association between e-commerce development and consumption growth is largest for the poorest quartile and declines as household income level moves up. While we have tried to control for as many factors as possible, some variables correlated with both the outcome variable and e-commerce indicator may have been omitted, which may bias the estimates of the e- commerce variable. Our findings imply only associations and not causality. More studies are needed to tease out the causal relationship between e-commerce development and consumption growth. Developing e-commerce requires not only Internet, but also many other factors, such as infrastructure and logistics services, skills and entrepreneurship, and an overarching enabling business environment facilitating online and offline business. China has been on a fast track of e- commerce development in recent years. With the right policies, further development of e- ··24    commerce can help accelerate consumption growth and narrow the regional gaps in e-commerce development, which in turn can lead to a more inclusive spatial development pattern. Due to data limitations, our study cannot separate total household consumption into online consumption and offline consumption, and therefore cannot examine the effects of e- commerce development on these two types of consumption separately. Studies on whether e- commerce results in additional offline consumption or online consumption substitutes for offline consumption will be interesting future research topics. Of equal if not more importance, studies on how e-commerce contributes to entrepreneurship and employment, and how this varies across sectors, regions, and individuals of different characteristics, will be crucial to understanding the role of e-commerce in the production and income aspects. In what areas and to what extent e- commerce adds to or substitutes for traditional offline business, and as a result, the net effect on the economy remains an important question for more empirical studies. References Agriculture and Agri-Food Canada (2017), E-grocery market in China, http://www.agr.gc.ca/eng/industry-markets-and-trade/international-agri-food-market- intelligence/asia/market-intelligence/e-grocery-market-in-china/?id=1504037238257. Baye, Michael R., and John Morgan, “Price Dispersion in the Lab and on the Internet: Theory and Evidence,” The RAND Journal of Economics 35, no. 3 (2004): 449. Baye, Michael R., John Morgan, and Patrick Scholten, “Price Dispersion in the Small and in the Large: Evidence From an Internet Price Comparison Site,” Journal of Industrial Economics 52, no. 4 (2004): 463–496. Blum, Bernardo S., and Avi Goldfarb, “Does the Internet Defy the Law of Gravity?,” Journal of International Economics 70, no. 2 (2006): 384–405. Brown, Jennifer R., Tanjim Hossain, and John Morgan, “Shrouded Attributes and Information Suppression: Evidence from the Field,” The Quarterly Journal of Economics 125, no. 2 (2010): 859–876. Brown, Jennifer R., and Austan Goolsbee, “Does the Internet Make Markets More Competitive?,” Journal of Political Economy 110, no. 3 (June 2002): 481–507. Brynjolfsson, Erik, Yu Hu, and Mohammad S. Rahman, “Battle of the Retail Channels: How Product Selection and Geography Drive Cross-Channel Competition,” Management Science 55, no. 11 (2009): 1755–1765. Cavallo, Alberto, “Are Online and Offline Prices Similar? Evidence from Large Multi-Channel Retailers,” American Economic Review 107, no. 1 (2017): 283–303. Choi, Changkyu, “The Effect of the Internet on Service Trade,” Economics Letters 109, no. 2 (2010): 102–104. Clay, Karen, Ramayya Krishnan, Eric Wolff, and Danny Fernandes, “Retail Strategies on the Web: Price and Non-Price Competition in the Online Book Industry,” The Journal of Industrial Economics 50 (2002): 351–367. ··25    Couture, Victor, Benjamin Faber, Yizhen Gu, and Lizhi Liu, "E-Commerce Integration and Economic Development: Evidence from China," NBER Working Paper No. 24384 (2018). Dai, Ruochen, and Xiaobo Zhang, “E-Commerce Expands the Bandwidth of Entrepreneurship,” manuscript, Peking University, 2017. Dolfen, Paul, Liran Einav, Peter J. Klenow, Benjamin Klopack, Jonathan D. Levin, Larry Levin, and Wayne Best, “Assessing the Gains from E-Commerce,” mimeo (2019) http://klenow.com/assessing-gains-ecommerce.pdf. Duch-Brown, Néstor, Lukasz Grzybowski, André Romahn, and FrankVerboven, “The Impact of Online Sales on Consumers and Firms: Evidence from Consumer Electronics,” International Journal of Industrial Organization 52, (2017): 30–62. Ellison, Glenn, and Sara Fisher Ellison, “Internet Retail Demand: Taxes, Geography and Online- Offline Competition,” National Bureau of Economic Research Working Paper Series 12242 (2006): 1–34. Ellison, Glenn, and Sara Fisher Ellison, “Match Quality, Search, and the Internet Market for Used Books,” manuscript, MIT, Febuary 2014: 1–34. Ellison, Glenn, and Sara Fisher Ellison, “Lessons About Markets from the Internet,” Journal of Economic Perspectives 19, no. 2 (2005): 139–158. Ellison, Glenn, and Sara Fisher Ellison, “Tax Sensitivity and Home State Preferences in Internet Purchasing,” American Economic Journal: Economic Policy 1, no. 12 (2009): 53–71. Ellison, Glenn, and Ellison Sara, “Search, Obfuscation, and Price Elasticities on the Internet,” Econometrica 77, no. 2 (2009): 427–452. Fan, Jingting, Lixin Tang, Weiming Zhu, and Ben Zou, “The Alibaba Effect: Spatial Consumption Inequality and the Welfare Gains from E-Commerce Jingting,” Working Paper, December (2016). Forman, Chris, Avi Goldfarb, and Shane Greenstein, “How Did Location Affect Adoption of the Commercial Internet? Global Village vs. Urban Leadership,” Journal of Urban Economics 58, no. 3 (2005): 389–420. Freund, Caroline L., and Diana Weinhold, “The Effect of the Internet on International Trade,” Journal of International Economics 62, no. 1 (2004): 171–189. Freund, Caroline L., and Diana Weinhold, “The Internet and International Trade in Services,” The American Economic Review 92, no. 2 (2002): 236–240. Garicano, Luis, and Kaplan, Steven N., "The Effects of Business-to-Business E-Commerce on Transaction Costs," Journal of Industrial Economics, Blackwell Publishing, 49(4) December 2001, pages 463-85. Goldmanis, Maris, Ali Hortaçsu, Chad Syverson, and Önsel Emre, “E-commerce and The Market Structure of Retail Industries,” The Economic Journal 120, no. 545 (2010): 651–682. Goolsbee, Austan, “Competition in the Computer Industry: Online versus Retail" Journal of Industrial Economics 49(4), December 2001: 487–499. Gorodnichenko, Yuriy, and Oleksandr Talavera, “Price Setting in Online Markets: Basic Facts, International Comparisons, and Cross-Border Integration,” American Economic Review 107, no.1 (2017): 249-282. Hortaçsu, Ali, F., Asís Martínez-Jerez, and Jason Douglas, “The Geography of Trade in Online Transactions: Evidence from eBay and MercadoLibre,” American Economic Journal: Microeconomics 1, no. 1 (2009): 53–74. Jensen, Robert, and Nolan Miller “Information, Demand and the Growth of Firms: Evidence from a Natural Experiment in India” Working paper, 2017. Jin, Ginger Zhe, and Andrew Kato, “Dividing Online and Offline: A Case Study,” Review of Economic Studies 74, no. 3 (2007): 981–1004. ··26    Kumar, Sameer, and Palo Petersen, "Impact of e‐commerce in lowering operational costs and raising customer satisfaction," Journal of Manufacturing Technology Management 17 no. 3 (2006), pp. 283–302, https://doi.org/10.1108/17410380610648263. Leamer, Edward E., and Michael Storper, “The Economic Geography of the Internet Age,” Journal of International Business Studies 32, no. 4 (2001): 641–65. Levin, Jonathan D., “The Economics of Internet Markets,” National Bureau of Economics Research Working Paper Series, no. 16852, March, 2011: 1–5. Lieber, Ethan, and Chad Syverson, “Online vs. Offline Competition,” in M.Peitz and J. Waldfogel (eds.), the Oxford Handbook of the Digital Economy, Oxford University Press, 2012. McKinsey (2013), China’s e-tail revolution: Online as a catalyst for growth. McKinsey & Company PwC's Entertainment and Media Outlook (2008), China's next retail disruption: End-to-end value chain digitalization. https://www.pwccn.com/en/retail-and-consumer/publications/global- consumer-insights-survey-2018-china-report.pdf. Radloff, L. S. (1977). The CES-D Scale: A Self-Report Depression Scale for Research in the General Population. Applied Psychological Measurement 1(3), 385–401, https://doi.org/10.1177/014662167700100306. Sinai, Todd, and Joel Waldfogel, “Geography and the Internet: Is the Internet a Substitute or a Complement for Cities?,” Journal of Urban Economics 56, no. 1 (2004): 1–24 Startz, Meredith, The Value of Face-to-Face: Search and Contracting Problems in Nigerian Trade (November 27, 2016). Available at SSRN: https://ssrn.com/abstract=3096685 or http://dx.doi.org/10.2139/ssrn.3096685. Zhang, Xiaobo, and Wu Zhu, “The Spatial Patterns of E-Commerce in China,” working paper, Peking University, 2015, 1–26. The World Bank Group, Digital Dividends World Bank Report 2016. ··27    Appendix 1. E-Commerce Index Construction To measure county-level e-commerce development, we constructed three sets of indicators: e- commerce penetration, e-commerce intensity, and e-commerce market size, following the two steps below: Step 1 Construct original indicators 1 E-commerce penetration The penetration index includes four indicators. Share of online sellers in population (%): 100 Share of online buyers in population (%): 100 Share of online sales in GDP (%): 100 Share of online purchase in GDP (%): 100 2 E-commerce intensity The intensity index includes four indicators. Per seller online sales (yuan): Per buyer online purchase (yuan): Per capita online sales (yuan): Per capita online purchase (yuan): 3 E-commerce market size The market size index includes four indicators. National share of online sales (%):（ ） 100 National share of online sellers (%): （ ） 100 National share of online purchase (%):（ ） 100 National share of online buyers (%): （ ） 100 ··28    Step 2 Normalize the original indicators to z-score We normalized each indicator to a z-score. The z-score is the difference between the raw value and the population mean divided by the population standard deviation. It is calculated using the formula below. raw value mean z standard deviation The z-score is the signed number of standard deviations by which the value of an observation is above or below the population mean value. It is an abstract value that is only used to represent the position in the population distribution of a data point. The mean of the z-score is 0 and the standard deviation of it is 1. The z-score is 0 for observed values at the mean level, positive above the mean and negative below the mean. The absolute value of the z-score measures the distance between a value and the mean value. The advantages of using the z-score are that it makes different indicators more comparable and that we could not infer the original value of each indicator. The constructed z-score indicators serve as proxy of e-commerce development level derived from the original raw data, including the gross merchandise values of online sales and online purchases, numbers of online sellers and online purchasers, and numbers of packages sent and received due to online sales and online purchases. Meanwhile, the constructed indicators preserve the confidentiality of the original data since the z-score reflects only the position of a data point in a distribution but not the value. If two observed values are from two identical distributions at the same relative position in each distribution, their z-score should be the same regardless of the difference in the raw value of the observation. Hence, the z-scores of the indicators cannot be used to infer the raw purchase/sales GMV, number of buyers/sellers, number of packages. ··29    Appendix 2. Robustness Check Table A.1. E-commerce development and total household expenditure per capita using alternative measures (1) (2) (3) (4) (5) (6) VARIABLES All Urban Rural East Central West Z-score of purchase/census pop in 0.303*** 0.247*** 0.718*** 0.267*** 0.313** 0.538*** 2013 (0.0671) (0.061) (0.098) (0.083) (0.119) (0.157) Z-score of buyers/census 0.189*** 0.150*** 0.455*** 0.127** 0.344*** 0.422*** population in 2013 (0.064) (0.057) (0.145) (0.057) (0.104) (0.117) Z-score of buyers share in the 0.208*** 0.175*** 0.734*** 0.179*** 0.660*** 0.664*** national market in 2013 (0.064) (0.051) (0.143) (0.040) (0.206) (0.194) Z-score of purchase share in 0.136*** 0.114*** 0.545*** 0.118*** 0.485** 0.649*** national market in 2013 (0.042) (0.033) (0.117) (0.026) (0.185) (0.199) Z-score of purchase over GDP in 0.275*** 0.269*** 0.266*** 0.431*** 0.171** 0.252** 2013 (0.067) (0.079) (0.088) (0.097) (0.084) (0.105) Z-score of per consumer purchase 0.126*** 0.149*** 0.093*** 0.148*** 0.112*** 0.022 in 2013 (0.021) (0.026) (0.029) (0.028) (0.036) (0.049) Observations 11,107 5,183 5,945 4,458 3,486 3,163 Note: The model of estimation is the same as Table 1, the dependent variable is the logarithmic form of household per capita consumption growth, and the variables of control include lagged log value of household expenditure per capita, average age/10, age squared/100, dependency ratio, urban-rural dummies, regional dummies, and year fixed effect except for gift and house repair and upgrading. For readability, we present only the results of interest - the coefficients of the z-score each measure in 2013. Fully results available upon request. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··30    Table A.2. E-commerce development and various categories of household expenditure per capita by region (Extended Model 1) (1) (2) (3) (4) (5) (6) VARIABLES All Urban Rural East Central West Total household expenditure 0.257*** 0.194*** 0.695*** 0.248*** 0.247* 0.538*** (0.067) (0.058) (0.103) (0.080) (0.126) (0.157) Observations 10,872 5,072 5,800 4,359 3,428 3,163 Adjusted R square 0.343 0.320 0.379 0.346 0.320 0.338 Cosmetics and beauty 0.404*** 0.339*** 0.945*** 0.462*** 0.289* 0.996*** (0.115) (0.112) (0.235) (0.151) (0.154) (0.334) Observations 10,764 5,023 5,741 4,312 3,399 3,053 Adjusted R square 0.344 0.367 0.326 0.330 0.361 0.361 Food 0.311*** 0.238*** 0.816*** 0.308*** 0.283** 0.523** (0.065) (0.056) (0.121) (0.079) (0.116) (0.229) Observations 10,720 4,993 5,727 4,279 3,396 3,045 Adjusted R square 0.345 0.343 0.360 0.359 0.309 0.386 Of which: Food at home 0.345*** 0.288*** 0.773*** 0.322*** 0.331*** 0.458** (0.071) (0.060) (0.160) (0.092) (0.116) (0.224) Observations 10,717 4,991 5,726 4,277 3,395 3,045 Adjusted R square 0.389 0.400 0.388 0.333 0.379 0.475 Of which: Dining out 0.424*** 0.286** 1.275*** 0.302* 0.561** 1.168** (0.142) (0.136) (0.272) (0.168) (0.240) (0.453) Observations 10,878 5,075 5,803 4,363 3,430 3,085 Adjusted R square 0.332 0.323 0.364 0.322 0.328 0.357 Clothes 0.217** 0.155* 0.671*** 0.196 0.243* 0.489* (0.095) (0.089) (0.171) (0.129) (0.128) (0.262) Observations 10,606 4,922 5,684 4,225 3,342 3,039 Adjusted R square 0.325 0.342 0.319 0.327 0.350 0.292 Utilities 0.156*** 0.153*** 0.251* 0.188*** 0.094* 0.417 (0.035) (0.035) (0.148) (0.045) (0.049) (0.309) Observations 10,658 4,945 5,713 4,282 3,363 3,013 Adjusted R square 0.305 0.294 0.317 0.312 0.297 0.325 Communications 0.189*** 0.150*** 0.523*** 0.149*** 0.223*** 0.402** (0.044) (0.038) (0.078) (0.052) (0.069) (0.190) Observations 10,546 4,902 5,644 4,212 3,344 2,990 Adjusted R square 0.337 0.353 0.335 0.308 0.387 0.313 Local transport 0.353*** 0.295*** 0.659*** 0.277** 0.402*** 0.307 (0.083) (0.078) (0.153) (0.114) (0.088) (0.291) Observations 10,490 4,865 5,625 4,188 3,323 2,979 Adjusted R square 0.306 0.318 0.299 0.290 0.321 0.318 Travel 0.713*** 0.549*** 1.495*** 0.668** 0.643*** 1.768*** (0.192) (0.177) (0.260) (0.284) (0.162) (0.508) ··31    Observations 10,667 4,950 5,717 4,287 3,369 3,011 Adjusted R square 0.297 0.289 0.340 0.288 0.276 0.355 Entertainment 0.550*** 0.409*** 1.159*** 0.461** 0.650*** 0.040 (0.145) (0.145) (0.339) (0.210) (0.164) (0.284) Observations 10,641 4,940 5,701 4,270 3,363 3,008 Adjusted R square 0.283 0.285 0.305 0.272 0.281 0.315 Automobiles 0.388*** 0.265** 1.434*** 0.302** 0.516** 0.929*** (0.124) (0.109) (0.289) (0.148) (0.218) (0.332) Observations 10,534 4,885 5,649 4,218 3,325 2,991 Adjusted R square 0.234 0.243 0.230 0.232 0.238 0.238 Other vehicles 0.154 0.119 0.869** 0.074 0.366 -0.001 (0.141) (0.129) (0.359) (0.129) (0.372) (0.586) Observations 10,626 4,936 5,690 4,263 3,357 3,006 Adjusted R square 0.396 0.400 0.395 0.399 0.395 0.397 Durable goods 0.320*** 0.184* 1.004** 0.317** 0.254 0.281 (0.120) (0.105) (0.396) (0.135) (0.193) (0.916) Observations 10,841 5,049 5,792 4,336 3,423 3,082 Adjusted R square 0.429 0.424 0.442 0.420 0.446 0.425 Medical 0.170** 0.130 0.442** 0.093 0.321** 0.471 (0.082) (0.094) (0.179) (0.117) (0.136) (0.410) Observations 10,733 5,004 5,729 4,279 3,395 3,059 Adjusted R square 0.377 0.366 0.396 0.388 0.382 0.358 Health and fitness 0.273** 0.143 0.544*** 0.211 0.257 0.529 (0.110) (0.102) (0.185) (0.142) (0.153) (0.325) Observations 10,844 5,055 5,789 4,349 3,422 3,073 Adjusted R square 0.303 0.286 0.367 0.302 0.320 0.290 Education -0.143 -0.116 0.093 -0.215** 0.079 -0.305 (0.114) (0.126) (0.268) (0.085) (0.203) (0.373) Observations 10,775 5,026 5,749 4,314 3,406 3,055 Adjusted R square 0.222 0.232 0.224 0.227 0.220 0.236 Home repair -0.089 -0.126 0.055 -0.068 -0.172 -0.359 (0.080) (0.085) (0.310) (0.089) (0.143) (0.629) Observations 10,666 4,961 5,705 4,282 3,372 3,012 Adjusted R square 0.450 0.433 0.470 0.476 0.456 0.408 Gifts -0.004 -0.005 0.390 -0.045 0.126 0.106 (0.101) (0.096) (0.273) (0.132) (0.127) (0.577) Observations 8,839 4,143 4,696 3,452 2,817 2,570 Adjusted R square 0.111 0.113 0.114 0.078 0.118 0.165 Note: Extended model with additional variables of control, including, lagged log value of household expenditure per capita of the consumption category in question, household average years of schooling, household gender ratio, average age/10, age squared/100, dependency ration, and urban-rural dummies. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··32    Table A.3. E-commerce development and various categories of household expenditure per capita by region (Extended Model 1) (1) (2) (3) (4) All bottom 25% 25%-75% top 25% Total household expenditure 0.257*** 0.305*** 0.143*** 0.053 (0.067) (0.102) (0.049) (0.048) Observations 10,872 2,984 5,627 2,232 Adjusted R square 0.343 0.416 0.440 0.446 Cosmetics and beauty 0.404*** 0.643** 0.197 0.164* (0.115) (0.246) (0.133) (0.084) Observations 10,764 2,956 5,578 2,201 Adjusted R square 0.344 0.330 0.370 0.423 Food 0.311*** 0.502*** 0.243*** 0.108** (0.065) (0.128) (0.058) (0.045) Observations 10,720 2,929 5,558 2,204 Adjusted R square 0.345 0.428 0.353 0.445 Of which: Food at home 0.345*** 0.538*** 0.279*** 0.177*** (0.071) (0.159) (0.071) (0.057) Observations 10,717 2,929 5,555 2,204 Adjusted R square 0.389 0.463 0.401 0.391 Of which: Dining out 0.424*** 0.169 0.144 0.111 (0.142) (0.214) (0.147) (0.118) Observations 10,878 2,987 5,630 2,232 Adjusted R square 0.332 0.455 0.341 0.380 Clothes 0.217** 0.190 0.036 -0.043 (0.095) (0.247) (0.085) (0.085) Observations 10,606 2,920 5,475 2,182 Adjusted R square 0.325 0.328 0.384 0.348 Utilities 0.156*** 0.305** 0.118*** 0.038 (0.035) (0.131) (0.039) (0.031) Observations 10,658 2,956 5,532 2,141 Adjusted R square 0.305 0.341 0.311 0.328 Communications 0.189*** 0.023 0.118** 0.087*** (0.044) (0.137) (0.049) (0.032) Observations 10,546 2,913 5,478 2,128 Adjusted R square 0.337 0.302 0.382 0.431 Local transport 0.353*** 0.048 0.148* 0.235* (0.083) (0.244) (0.080) (0.119) Observations 10,490 2,911 5,437 2,113 Adjusted R square 0.306 0.338 0.327 0.310 Travel 0.713*** 0.811** 0.371** 0.411** (0.192) (0.322) (0.147) (0.190) ··33    Observations 10,667 2,957 5,536 2,145 Adjusted R square 0.297 0.437 0.346 0.283 Entertainment 0.550*** 0.159 0.335* 0.341*** (0.145) (0.146) (0.179) (0.107) Observations 10,641 2,952 5,516 2,144 Adjusted R square 0.283 0.315 0.303 0.326 Automobiles 0.388*** 0.229 0.132 0.330** (0.124) (0.177) (0.154) (0.144) Observations 10,534 2,935 5,465 2,106 Adjusted R square 0.234 0.333 0.233 0.243 Other vehicles 0.154 -0.048 0.025 0.022 (0.141) (0.295) (0.196) (0.119) Observations 10,626 2,947 5,507 2,144 Adjusted R square 0.396 0.412 0.403 0.396 Durable goods 0.320*** -0.225 0.024 0.062 (0.120) (0.358) (0.122) (0.128) Observations 10,841 2,984 5,610 2,218 Adjusted R square 0.429 0.488 0.445 0.409 Medical 0.170** 0.548** -0.041 0.192 (0.082) (0.247) (0.120) (0.158) Observations 10,733 2,947 5,550 2,207 Adjusted R square 0.377 0.379 0.384 0.382 Health and fitness 0.273** 0.322* -0.044 0.057 (0.110) (0.164) (0.104) (0.107) Observations 10,844 2,980 5,615 2,220 Adjusted R square 0.303 0.434 0.335 0.287 Education -0.143 -0.552** -0.124 0.040 (0.114) (0.228) (0.113) (0.223) Observations 10,775 2,954 5,573 2,219 Adjusted R square 0.222 0.230 0.219 0.279 Home repair -0.089 -0.144 -0.112 -0.400*** (0.080) (0.328) (0.120) (0.131) Observations 10,666 2,959 5,531 2,147 Adjusted R square 0.450 0.435 0.462 0.453 Gifts -0.004 0.437 0.002 -0.310** (0.101) (0.413) (0.115) (0.145) Observations 8,839 2,258 4,701 1,857 Adjusted R square 0.111 0.140 0.117 0.160 Note: Extended model with additional variables of control, including, lagged log value of household expenditure per capita of the consumption category in question, household average years of schooling, household gender ratio, average age/10, age squared/100, dependency ration, and urban-rural dummies. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··34    Table A.4. E-commerce development and various categories of household expenditure per capita by region (Extended Model 2) (1) (2) (3) (4) (5) (6) VARIABLES All Urban Rural East Central West Total household expenditure 0.245*** 0.176*** 0.722*** 0.231*** 0.234** 0.538*** (0.062) (0.052) (0.096) (0.078) (0.107) (0.157) Observations 9,502 4,402 5,100 3,808 3,018 3,163 Adjusted R square 0.341 0.314 0.383 0.336 0.325 0.338 Cosmetics and beauty 0.406*** 0.327*** 1.011*** 0.419*** 0.311* 1.259*** (0.125) (0.118) (0.248) (0.155) (0.181) (0.302) Observations 9,422 4,366 5,056 3,772 2,997 2,653 Adjusted R square 0.342 0.367 0.323 0.317 0.368 0.362 Food 0.289*** 0.219*** 0.764*** 0.302*** 0.233** 0.479* (0.063) (0.053) (0.122) (0.079) (0.099) (0.243) Observations 9,398 4,346 5,052 3,751 2,998 2,649 Adjusted R square 0.326 0.321 0.341 0.333 0.302 0.360 Of which: Food at home 0.342*** 0.279*** 0.814*** 0.338*** 0.291** 0.453* (0.067) (0.059) (0.126) (0.086) (0.108) (0.257) Observations 9,395 4,344 5,051 3,749 2,997 2,649 Adjusted R square 0.375 0.385 0.375 0.316 0.372 0.459 Of which: Dining out 0.349** 0.214 1.202*** 0.234 0.464* 1.165** (0.146) (0.132) (0.265) (0.173) (0.240) (0.496) Observations 9,503 4,402 5,101 3,808 3,019 2,676 Adjusted R square 0.329 0.321 0.359 0.308 0.331 0.359 Clothes 0.211** 0.144 0.738*** 0.195 0.233 0.502* (0.103) (0.096) (0.178) (0.137) (0.146) (0.289) Observations 9,297 4,285 5,012 3,704 2,951 2,642 Adjusted R square 0.321 0.341 0.312 0.312 0.349 0.304 Utilities 0.149*** 0.141*** 0.321** 0.190*** 0.076* 0.486 (0.034) (0.034) (0.141) (0.044) (0.041) (0.307) Observations 9,469 4,376 5,093 3,787 3,012 2,670 Adjusted R square 0.298 0.284 0.313 0.311 0.292 0.309 Communications 0.190*** 0.154*** 0.547*** 0.148** 0.233*** 0.375** (0.045) (0.040) (0.084) (0.057) (0.061) (0.177) Observations 9,382 4,338 5,044 3,734 2,996 2,652 Adjusted R square 0.341 0.364 0.334 0.303 0.397 0.324 Local transport 0.364*** 0.297*** 0.807*** 0.300** 0.410*** 0.281 (0.083) (0.077) (0.165) (0.117) (0.083) (0.299) Observations 9,335 4,308 5,027 3,708 2,979 2,648 Adjusted R square 0.299 0.307 0.296 0.285 0.315 0.303 Travel 0.704*** 0.534*** 1.609*** 0.638** 0.657*** 1.888*** (0.198) (0.184) (0.257) (0.285) (0.189) (0.567) Observations 9,478 4,382 5,096 3,794 3,014 2,670 Adjusted R square 0.296 0.291 0.333 0.286 0.279 0.350 Entertainment 0.528*** 0.379** 1.158*** 0.398* 0.687*** 0.106 (0.151) (0.147) (0.348) (0.212) (0.186) (0.287) Observations 9,455 4,371 5,084 3,778 3,011 2,666 Adjusted R square 0.289 0.291 0.314 0.270 0.292 0.325 Automobiles 0.384*** 0.247** 1.600*** 0.291* 0.505** 0.979*** ··35    (0.133) (0.119) (0.339) (0.169) (0.222) (0.329) Observations 9,370 4,328 5,042 3,733 2,981 2,656 Adjusted R square 0.233 0.237 0.238 0.229 0.234 0.248 Other vehicles 0.179 0.123 0.994** 0.101 0.352 0.097 (0.137) (0.125) (0.387) (0.130) (0.348) (0.677) Observations 9,448 4,371 5,077 3,773 3,008 2,667 Adjusted R square 0.393 0.394 0.396 0.391 0.397 0.392 Durable goods 0.348*** 0.201* 1.039** 0.315** 0.314 0.298 (0.130) (0.120) (0.400) (0.145) (0.224) (0.871) Observations 9,476 4,381 5,095 3,787 3,015 2,674 Adjusted R square 0.429 0.422 0.444 0.420 0.446 0.422 Medical 0.146* 0.095 0.459*** 0.069 0.281* 0.458 (0.082) (0.096) (0.171) (0.110) (0.143) (0.410) Observations 9,395 4,350 5,045 3,748 2,989 2,658 Adjusted R square 0.362 0.350 0.385 0.375 0.370 0.339 Health and fitness 0.222** 0.095 0.514** 0.100 0.322* 0.610* (0.107) (0.099) (0.215) (0.132) (0.165) (0.353) Observations 9,478 4,384 5,094 3,798 3,014 2,666 Adjusted R square 0.299 0.278 0.371 0.294 0.313 0.301 Education -0.142 -0.127 0.167 -0.232*** 0.087 0.082 (0.116) (0.130) (0.269) (0.083) (0.201) (0.396) Observations 9,417 4,359 5,058 3,767 2,997 2,653 Adjusted R square 0.217 0.228 0.219 0.221 0.210 0.245 Home repair -0.065 -0.153* 0.393 -0.082 -0.186 0.088 (0.082) (0.084) (0.349) (0.103) (0.134) (0.648) Observations 9,470 4,386 5,084 3,785 3,017 2,668 Adjusted R square 0.452 0.437 0.469 0.483 0.445 0.420 Gifts -0.013 -0.031 0.545** -0.050 0.115 0.170 (0.101) (0.098) (0.243) (0.137) (0.113) (0.495) Observations 7,802 3,626 4,176 3,039 2,525 2,238 Adjusted R square 0.111 0.113 0.116 0.079 0.120 0.167 Note: Extended model with additional variables of control, including the lagged household per capita income growth, lagged log value of household expenditure per capita of the consumption category in question, household average years of schooling household gender ratio, average age/10, age squared/100, dependency ration, and urban-rural dummies. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··36    Table A.5. E-commerce development and various categories of household expenditure per capita by household consumption level (Extended model 2) (1) (2) (3) (4) All bottom 25% 25%-75% top 25% Total household expenditure 0.245*** 0.371*** 0.142*** 0.039 (0.062) (0.078) (0.047) (0.050) Observations 9,502 2,590 4,945 1,946 Adjusted R square 0.341 0.407 0.431 0.456 Cosmetics and beauty 0.406*** 0.540** 0.214 0.177* (0.125) (0.272) (0.146) (0.098) Observations 9,422 2,569 4,905 1,927 Adjusted R square 0.342 0.322 0.370 0.425 Food 0.289*** 0.447*** 0.226*** 0.104** (0.063) (0.134) (0.055) (0.043) Observations 9,398 2,551 4,895 1,931 Adjusted R square 0.326 0.395 0.331 0.457 Of which: Food at home 0.342*** 0.502*** 0.267*** 0.203*** (0.067) (0.170) (0.071) (0.056) Observations 9,395 2,551 4,892 1,931 Adjusted R square 0.375 0.447 0.384 0.382 Of which: Dining out 0.349** 0.155 0.097 0.034 (0.146) (0.245) (0.155) (0.118) Observations 9,503 2,590 4,946 1,946 Adjusted R square 0.329 0.458 0.335 0.364 Clothes 0.211** 0.187 0.062 -0.054 (0.103) (0.267) (0.096) (0.093) Observations 9,297 2,536 4,833 1,907 Adjusted R square 0.321 0.325 0.374 0.353 Utilities 0.149*** 0.337** 0.105*** 0.051 (0.034) (0.129) (0.036) (0.032) Observations 9,469 2,584 4,935 1,929 Adjusted R square 0.298 0.321 0.303 0.349 Communications 0.190*** 0.029 0.123** 0.094*** (0.045) (0.135) (0.052) (0.034) Observations 9,382 2,557 4,888 1,918 Adjusted R square 0.341 0.312 0.378 0.434 Local transport 0.364*** -0.053 0.181** 0.261** (0.083) (0.263) (0.086) (0.120) Observations 9,335 2,555 4,855 1,904 Adjusted R square 0.299 0.339 0.315 0.302 Travel 0.704*** 0.966** 0.376** 0.393** (0.198) (0.380) (0.162) (0.186) Observations 9,478 2,586 4,936 1,935 Adjusted R square 0.296 0.420 0.350 0.277 Entertainment 0.528*** 0.244 0.290 0.338*** (0.151) (0.175) (0.180) (0.115) Observations 9,455 2,580 4,921 1,933 Adjusted R square 0.289 0.332 0.308 0.321 Automobiles 0.384*** 0.297 0.170 0.298* ··37    (0.133) (0.203) (0.164) (0.162) Observations 9,370 2,567 4,881 1,902 Adjusted R square 0.233 0.329 0.233 0.238 Other vehicles 0.179 0.279 0.002 0.051 (0.137) (0.309) (0.205) (0.117) Observations 9,448 2,578 4,917 1,933 Adjusted R square 0.393 0.406 0.405 0.388 Durable goods 0.348*** -0.113 0.075 0.057 (0.130) (0.428) (0.149) (0.134) Observations 9,476 2,588 4,930 1,937 Adjusted R square 0.429 0.486 0.444 0.403 Medical 0.146* 0.584** -0.049 0.169 (0.082) (0.290) (0.113) (0.145) Observations 9,395 2,559 4,887 1,928 Adjusted R square 0.362 0.381 0.363 0.367 Health and fitness 0.222** 0.151 -0.047 -0.012 (0.107) (0.144) (0.107) (0.114) Observations 9,478 2,587 4,934 1,936 Adjusted R square 0.299 0.419 0.334 0.284 Education -0.142 -0.414* -0.158 0.086 (0.116) (0.241) (0.119) (0.203) Observations 9,417 2,563 4,899 1,934 Adjusted R square 0.217 0.226 0.215 0.271 Home repair -0.065 -0.064 -0.010 -0.447*** (0.082) (0.383) (0.130) (0.131) Observations 9,470 2,586 4,931 1,932 Adjusted R square 0.452 0.434 0.463 0.460 Gifts -0.013 0.562 -0.009 -0.306** (0.101) (0.387) (0.119) (0.154) Observations 7,802 1,983 4,172 1,631 Adjusted R square 0.111 0.142 0.117 0.162 Note: Extended model with additional variables of control, including the lagged log household per capita income growth, lagged log value of household expenditure per capita of the consumption category in question, household average years of schooling, household gender ratio, average age squared/100, dependency ratio, and urban-rural dummies. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··38    Table A.6. E-commerce development and various categories of household expenditure per capita by region (Extended Model 3) (1) (2) (3) (4) (5) (6) VARIABLES All Urban Rural East Central West Total household expenditure 0.210*** 0.151*** 0.623*** 0.207*** 0.189 0.538*** (0.061) (0.054) (0.103) (0.067) (0.124) (0.157) Observations 10,171 4,740 5,431 4,049 3,185 3,163 Adjusted R square 0.366 0.345 0.395 0.374 0.347 0.338 Cosmetics and beauty 0.328*** 0.256** 0.845*** 0.387*** 0.221 0.879*** (0.110) (0.107) (0.226) (0.128) (0.165) (0.320) Observations 10,069 4,695 5,374 4,006 3,157 2,906 Adjusted R square 0.349 0.376 0.326 0.338 0.362 0.367 Food 0.268*** 0.205*** 0.718*** 0.280*** 0.218* 0.444** (0.059) (0.052) (0.124) (0.070) (0.113) (0.218) Observations 10,037 4,673 5,364 3,982 3,155 2,900 Adjusted R square 0.358 0.360 0.367 0.373 0.337 0.384 Of which: Food at home 0.289*** 0.245*** 0.639*** 0.289*** 0.240** 0.362* (0.063) (0.055) (0.165) (0.086) (0.107) (0.214) Observations 10,034 4,671 5,363 3,980 3,154 2,900 Adjusted R square 0.399 0.416 0.390 0.340 0.410 0.470 Of which: Dining out 0.330** 0.171 1.192*** 0.207 0.444* 1.092** (0.134) (0.127) (0.266) (0.151) (0.241) (0.493) Observations 10,176 4,742 5,434 4,052 3,187 2,937 Adjusted R square 0.337 0.328 0.370 0.326 0.333 0.362 Clothes 0.137 0.079 0.513*** 0.128 0.155 0.304 (0.087) (0.082) (0.169) (0.115) (0.131) (0.254) Observations 9,924 4,603 5,321 3,926 3,104 2,894 Adjusted R square 0.338 0.359 0.327 0.346 0.359 0.301 Utilities 0.130*** 0.130*** 0.191 0.174*** 0.037 0.408 (0.034) (0.037) (0.141) (0.041) (0.042) (0.307) Observations 9,972 4,623 5,349 3,979 3,125 2,868 Adjusted R square 0.305 0.302 0.311 0.313 0.303 0.319 Communications 0.156*** 0.117*** 0.481*** 0.126** 0.175** 0.334 (0.041) (0.034) (0.080) (0.049) (0.066) (0.200) Observations 9,868 4,582 5,286 3,915 3,108 2,845 Adjusted R square 0.348 0.373 0.338 0.322 0.407 0.311 Local transport 0.292*** 0.235*** 0.527*** 0.223** 0.319*** 0.137 (0.075) (0.068) (0.160) (0.102) (0.090) (0.289) Observations 9,812 4,546 5,266 3,886 3,087 2,839 Adjusted R square 0.319 0.335 0.309 0.305 0.335 0.327 Travel 0.579*** 0.395*** 1.270*** 0.518** 0.489*** 1.697*** (0.164) (0.146) (0.200) (0.239) (0.153) (0.484) ··39    Observations 9,978 4,626 5,352 3,982 3,131 2,865 Adjusted R square 0.304 0.300 0.342 0.301 0.280 0.352 Entertainment 0.519*** 0.394*** 1.020*** 0.417** 0.618*** 0.036 (0.139) (0.141) (0.310) (0.194) (0.181) (0.300) Observations 9,955 4,616 5,339 3,967 3,126 2,862 Adjusted R square 0.284 0.288 0.299 0.271 0.281 0.319 Automobiles 0.311*** 0.204* 1.234*** 0.210 0.423* 0.846*** (0.118) (0.106) (0.273) (0.135) (0.219) (0.295) Observations 9,850 4,565 5,285 3,912 3,088 2,850 Adjusted R square 0.241 0.250 0.238 0.240 0.239 0.252 Other vehicles 0.085 0.062 0.728** 0.007 0.288 -0.061 (0.146) (0.139) (0.364) (0.132) (0.395) (0.561) Observations 9,940 4,614 5,326 3,960 3,119 2,861 Adjusted R square 0.397 0.403 0.395 0.402 0.399 0.392 Durable goods 0.187 0.052 0.853** 0.176 0.121 0.181 (0.118) (0.110) (0.389) (0.121) (0.188) (0.929) Observations 10,141 4,718 5,423 4,027 3,180 2,934 Adjusted R square 0.435 0.432 0.443 0.423 0.455 0.429 Medical 0.136 0.097 0.442** 0.040 0.291** 0.515 (0.089) (0.101) (0.195) (0.134) (0.142) (0.388) Observations 10,045 4,679 5,366 3,978 3,154 2,913 Adjusted R square 0.376 0.369 0.389 0.383 0.383 0.359 Health and fitness 0.183* 0.055 0.345* 0.118 0.179 0.383 (0.098) (0.093) (0.195) (0.116) (0.163) (0.307) Observations 10,144 4,722 5,422 4,040 3,179 2,925 Adjusted R square 0.309 0.293 0.372 0.304 0.330 0.300 Education -0.142 -0.131 0.174 -0.231*** 0.035 -0.089 (0.114) (0.126) (0.274) (0.086) (0.196) (0.358) Observations 10,081 4,695 5,386 4,007 3,165 2,909 Adjusted R square 0.221 0.231 0.220 0.226 0.217 0.236 Home repair -0.124 -0.156* 0.033 -0.114 -0.263 -0.213 (0.087) (0.091) (0.342) (0.081) (0.162) (0.660) Observations 9,975 4,633 5,342 3,975 3,134 2,866 Adjusted R square 0.452 0.434 0.471 0.483 0.455 0.407 Gifts -0.062 -0.054 0.227 -0.115 0.053 0.011 (0.100) (0.098) (0.275) (0.134) (0.133) (0.608) Observations 8,279 3,870 4,409 3,214 2,623 2,442 Adjusted R square 0.111 0.108 0.118 0.078 0.122 0.165 Note: Alternative extended model, with additional variables of control, including the lagged household per capita income, legged log of household expenditure per capita of the consumption category in question, household average years of schooling household gender ratio, average age, age squared/100, dependency ration, and urban-rural dummies. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··40    Table A.7. E-commerce development and various categories of household expenditure per capita by household consumption level (Extended model 3) (1) (2) (3) (4) All bottom 25% 25%-75% top 25% Total household expenditure 0.210*** 0.289*** 0.114** 0.047 (0.061) (0.105) (0.046) (0.043) 10,171 2,770 5,287 2,088 0.366 0.427 0.449 0.458 Cosmetics and beauty 0.328*** 0.555** 0.176 0.128* (0.110) (0.251) (0.137) (0.071) Observations 10,069 2,743 5,239 2,061 Adjusted R square 0.349 0.332 0.365 0.436 Food 0.268*** 0.466*** 0.227*** 0.094** (0.059) (0.131) (0.055) (0.044) Observations 10,034 2,720 5,225 2,063 Adjusted R square 0.399 0.464 0.406 0.400 Of which: Food at home 0.289*** 0.492*** 0.264*** 0.144** (0.063) (0.167) (0.069) (0.055) Observations 10,034 2,018 5,116 2,877 Adjusted R square 0.399 0.464 0.366 0.410 Of which: Dining out 0.330** 0.141 0.081 0.061 (0.134) (0.220) (0.135) (0.123) Observations 9,924 2,712 5,144 2,042 Adjusted R square 0.338 0.329 0.393 0.376 Clothes 0.137 0.165 -0.002 -0.066 (0.087) (0.248) (0.090) (0.075) Observations 9,924 2,061 5,178 2,911 Adjusted R square 0.338 0.479 0.362 0.343 Utilities 0.130*** 0.279** 0.111*** 0.023 (0.034) (0.126) (0.040) (0.030) Observations 9,972 2,743 5,198 2,005 Adjusted R square 0.305 0.330 0.313 0.328 Communications 0.156*** 0.033 0.106** 0.081*** (0.041) (0.131) (0.049) (0.029) Observations 9,868 2,704 5,148 1,991 Adjusted R square 0.348 0.294 0.398 0.436 Local transport 0.292*** 0.030 0.121 0.220** (0.075) (0.243) (0.082) (0.110) Observations 9,812 2,703 5,105 1,978 Adjusted R square 0.319 0.344 0.335 0.325 Travel 0.579*** 0.771** 0.299** 0.322* (0.164) (0.327) (0.126) (0.170) ··41    Observations 9,978 2,745 5,199 2,008 Adjusted R square 0.304 0.450 0.345 0.288 Entertainment 0.519*** 0.190 0.298* 0.374*** (0.139) (0.144) (0.178) (0.098) Observations 9,850 2,722 5,132 1,971 Adjusted R square 0.241 0.336 0.240 0.248 Automobiles 0.311*** 0.199 0.094 0.307** (0.118) (0.172) (0.150) (0.125) Observations 9,850 2,722 5,132 1,971 Adjusted R square 0.241 0.336 0.240 0.248 Other vehicles 0.085 -0.129 0.046 -0.030 (0.146) (0.288) (0.206) (0.131) Observations 9,940 2,735 5,173 2,007 Adjusted R square 0.397 0.416 0.401 0.399 Durable goods 0.187 -0.275 -0.019 0.010 (0.118) (0.380) (0.120) (0.132) Observations 10,141 2,770 5,270 2,075 Adjusted R square 0.435 0.496 0.446 0.412 Medical 0.136 0.498* -0.051 0.181 (0.089) (0.257) (0.118) (0.172) Observations 10,045 2,737 5,216 2,066 Adjusted R square 0.376 0.367 0.386 0.385 Health and fitness 0.183* 0.229* -0.059 -0.007 (0.098) (0.135) (0.107) (0.111) Observations 10,144 2,766 5,276 2,076 Adjusted R square 0.309 0.424 0.338 0.294 Education -0.142 -0.410* -0.160 0.024 (0.114) (0.213) (0.123) (0.202) Observations 10,081 2,744 5,235 2,076 Adjusted R square 0.221 0.229 0.217 0.275 Home repair -0.124 -0.090 -0.169 -0.325** (0.087) (0.341) (0.128) (0.136) Observations 9,975 2,746 5,196 2,007 Adjusted R square 0.452 0.435 0.465 0.451 Gifts -0.062 0.505 -0.014 -0.376** (0.100) (0.379) (0.115) (0.152) Observations 8,279 2,094 4,425 1,738 Adjusted R square 0.111 0.133 0.119 0.150 Note: Alternative extended model, with additional variables of control, including the lagged log household per capita, income per capita lagged log value of household expenditure per capita of the consumption category in question, household average years of schooling, household gender ratio, average age squared/100, dependency ratio, and urban- rural dummies. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··42    Table A.8. E-commerce development and various categories of household expenditure per capita by region (Extended Model 4) (1) (2) (3) (4) (5) (6) VARIABLES All Urban Rural East Central West Total household expenditure 0.226*** 0.155*** 0.692*** 0.170** 0.238* 0.538*** (0.064) (0.056) (0.105) (0.069) (0.125) (0.157) Observations 10,701 5,072 5,629 4,253 3,428 3,163 Adjusted R square 0.344 0.324 0.378 0.351 0.320 0.338 Cosmetics and beauty 0.353*** 0.280*** 0.914*** 0.396** 0.258* 1.004*** (0.113) (0.106) (0.251) (0.150) (0.150) (0.328) Observations 10,597 5,023 5,574 4,210 3,399 2,988 Adjusted R square 0.343 0.368 0.322 0.332 0.361 0.353 Food 0.283*** 0.229*** 0.737*** 0.238*** 0.269** 0.479** (0.063) (0.056) (0.121) (0.070) (0.117) (0.217) Observations 10,550 4,993 5,557 4,174 3,396 2,980 Adjusted R square 0.347 0.343 0.365 0.366 0.310 0.390 Of which: Food at home 0.304*** 0.266*** 0.682*** 0.240*** 0.309** 0.425* (0.067) (0.058) (0.164) (0.084) (0.115) (0.224) Observations 10,547 4,991 5,556 4,172 3,395 2,980 Adjusted R square 0.390 0.400 0.390 0.339 0.380 0.476 Of which: Dining out 0.424*** 0.267* 1.258*** 0.210 0.589** 1.137** (0.148) (0.147) (0.273) (0.161) (0.245) (0.446) Observations 10,707 5,075 5,632 4,257 3,430 3,020 Adjusted R square 0.331 0.322 0.361 0.324 0.329 0.353 Clothes 0.192** 0.125 0.634*** 0.145 0.232* 0.433* (0.094) (0.091) (0.186) (0.131) (0.129) (0.251) Observations 10,437 4,922 5,515 4,121 3,342 2,974 Adjusted R square 0.326 0.342 0.319 0.327 0.350 0.295 Utilities 0.171*** 0.138*** 0.321** 0.169*** 0.122** 0.432 (0.041) (0.036) (0.153) (0.048) (0.057) (0.281) Observations 10,493 4,945 5,548 4,178 3,363 2,952 Adjusted R square 0.308 0.295 0.329 0.313 0.300 0.337 Communications 0.177*** 0.132*** 0.525*** 0.119** 0.210*** 0.396** (0.044) (0.038) (0.086) (0.052) (0.066) (0.178) Observations 10,383 4,902 5,481 4,110 3,344 2,929 Adjusted R square 0.337 0.354 0.334 0.308 0.387 0.316 Local transport 0.332*** 0.272*** 0.601*** 0.217* 0.407*** 0.215 (0.081) (0.077) (0.160) (0.111) (0.090) (0.289) Observations 10,327 4,865 5,462 4,086 3,323 2,918 Adjusted R square 0.306 0.318 0.298 0.290 0.321 0.321 Travel 0.650*** 0.465*** 1.423*** 0.496* 0.626*** 1.743*** (0.187) (0.169) (0.273) (0.272) (0.156) (0.515) ··43    Observations 10,502 4,950 5,552 4,183 3,369 2,950 Adjusted R square 0.297 0.291 0.337 0.291 0.276 0.355 Entertainment 0.495*** 0.342** 1.065*** 0.383* 0.600*** 0.016 (0.143) (0.142) (0.356) (0.215) (0.156) (0.291) Observations 10,477 4,940 5,537 4,167 3,363 2,947 Adjusted R square 0.284 0.287 0.305 0.275 0.282 0.315 Automobiles 0.310** 0.158 1.465*** 0.213 0.458** 0.905*** (0.120) (0.105) (0.305) (0.160) (0.196) (0.318) Observations 10,372 4,885 5,487 4,116 3,325 2,931 Adjusted R square 0.235 0.245 0.230 0.231 0.239 0.245 Other vehicles 0.057 -0.023 0.818** -0.103 0.304 -0.022 (0.143) (0.128) (0.387) (0.123) (0.372) (0.591) Observations 10,461 4,936 5,525 4,159 3,357 2,945 Adjusted R square 0.397 0.403 0.396 0.402 0.396 0.398 Durable goods 0.178 0.025 0.929** 0.125 0.172 0.220 (0.114) (0.101) (0.426) (0.112) (0.195) (0.917) Observations 10,670 5,049 5,621 4,230 3,423 3,017 Adjusted R square 0.430 0.427 0.439 0.422 0.447 0.422 Medical 0.163** 0.124 0.407** 0.085 0.327** 0.429 (0.082) (0.097) (0.192) (0.135) (0.123) (0.391) Observations 10,562 5,004 5,558 4,173 3,395 2,994 Adjusted R square 0.378 0.366 0.397 0.387 0.382 0.359 Health and fitness 0.221** 0.059 0.563*** 0.148 0.217 0.519 (0.106) (0.093) (0.200) (0.138) (0.144) (0.327) Observations 10,673 5,055 5,618 4,243 3,422 3,008 Adjusted R square 0.303 0.288 0.366 0.301 0.321 0.289 Education -0.151 -0.156 0.095 -0.185** 0.010 -0.248 (0.117) (0.130) (0.274) (0.085) (0.201) (0.369) Observations 10,604 5,026 5,578 4,208 3,406 2,990 Adjusted R square 0.221 0.232 0.222 0.227 0.221 0.232 Home repair -0.121 -0.202** 0.077 -0.121 -0.209 -0.458 (0.091) (0.095) (0.319) (0.095) (0.158) (0.616) Observations 10,505 4,961 5,544 4,182 3,372 2,951 Adjusted R square 0.449 0.433 0.468 0.473 0.456 0.409 Gifts -0.028 -0.023 0.353 -0.031 0.074 0.121 (0.107) (0.107) (0.286) (0.156) (0.128) (0.565) Observations 8,716 4,143 4,573 3,377 2,817 2,522 Adjusted R square 0.112 0.113 0.116 0.077 0.121 0.171 Note: Alternative extended model, with additional variables of control, including provincial CPI of the corresponding consumption category, lagged value of log household per capita in consumption category in question, household average years of schooling household gender ratio, average age, age squared/100, dependency ration, and urban-rural dummies. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··44    Table A.9. E-commerce development and various categories of household expenditure per capita by household consumption level (Extended model 4) (1) (2) (3) (4) All bottom 25% 25%-75% top 25% Total household expenditure 0.226*** 0.287*** 0.131*** 0.043 (0.064) (0.100) (0.048) (0.051) Observations 10,701 2,959 5,547 2,166 Adjusted R square 0.344 0.417 0.439 0.442 Cosmetics and beauty 0.353*** 0.570** 0.194 0.141 (0.113) (0.248) (0.135) (0.089) Observations 10,597 2,931 5,500 2,137 Adjusted R square 0.343 0.330 0.367 0.425 Food 0.283*** 0.442*** 0.236*** 0.098** (0.063) (0.129) (0.058) (0.048) Observations 10,550 2,904 5,478 2,139 Adjusted R square 0.347 0.428 0.356 0.447 Of which: Food at home 0.304*** 0.456*** 0.264*** 0.149** (0.067) (0.154) (0.070) (0.060) Observations 10,547 2,904 5,475 2,139 Adjusted R square 0.390 0.465 0.401 0.392 Of which: Dining out 0.424*** 0.191 0.172 0.143 (0.148) (0.218) (0.147) (0.137) Observations 10,707 2,962 5,550 2,166 Adjusted R square 0.331 0.453 0.341 0.379 Clothes 0.192** 0.204 0.036 -0.068 (0.094) (0.254) (0.086) (0.089) Observations 10,437 2,895 5,396 2,117 Adjusted R square 0.326 0.330 0.386 0.348 Utilities 0.171*** 0.347** 0.145*** 0.027 (0.041) (0.134) (0.047) (0.033) Observations 10,493 2,931 5,457 2,076 Adjusted R square 0.308 0.345 0.313 0.333 Communications 0.177*** 0.049 0.116** 0.071** (0.044) (0.138) (0.049) (0.035) Observations 10,383 2,888 5,404 2,064 Adjusted R square 0.337 0.302 0.384 0.432 Local transport 0.332*** 0.045 0.139* 0.233* (0.081) (0.247) (0.079) (0.128) Observations 10,327 2,886 5,362 2,050 Adjusted R square 0.306 0.340 0.327 0.309 Travel 0.650*** 0.841** 0.358** 0.305* (0.187) (0.325) (0.146) (0.182) ··45    Observations 10,502 2,932 5,461 2,080 Adjusted R square 0.297 0.437 0.347 0.282 Entertainment 0.495*** 0.168 0.279 0.328*** (0.143) (0.147) (0.174) (0.109) Observations 10,477 2,927 5,441 2,080 Adjusted R square 0.284 0.313 0.304 0.327 Automobiles 0.310** 0.199 0.063 0.308** (0.120) (0.175) (0.151) (0.155) Observations 10,372 2,910 5,391 2,043 Adjusted R square 0.235 0.333 0.234 0.243 Other vehicles 0.057 -0.102 -0.050 -0.073 (0.143) (0.302) (0.199) (0.111) Observations 10,461 2,922 5,432 2,079 Adjusted R square 0.397 0.413 0.404 0.400 Durable goods 0.178 -0.256 -0.072 -0.063 (0.114) (0.373) (0.128) (0.127) Observations 10,670 2,959 5,530 2,152 Adjusted R square 0.430 0.489 0.445 0.409 Medical 0.163** 0.538** -0.083 0.274 (0.082) (0.237) (0.116) (0.173) Observations 10,562 2,922 5,470 2,141 Adjusted R square 0.378 0.378 0.385 0.383 Health and fitness 0.221** 0.328** -0.070 0.024 (0.106) (0.163) (0.108) (0.106) Observations 10,673 2,955 5,535 2,154 Adjusted R square 0.303 0.432 0.337 0.284 Education -0.151 -0.578** -0.137 0.028 (0.117) (0.233) (0.112) (0.243) Observations 10,604 2,929 5,493 2,153 Adjusted R square 0.221 0.230 0.217 0.278 - Home repair -0.121 -0.128 -0.140 0.436*** (0.091) (0.336) (0.128) (0.135) Observations 10,505 2,934 5,456 2,086 Adjusted R square 0.449 0.434 0.463 0.446 Gifts -0.028 0.357 -0.005 -0.299* (0.107) (0.419) (0.118) (0.159) Observations 8,716 2,240 4,643 1,810 Adjusted R square 0.112 0.142 0.117 0.160 Note: Alternative extended model with additional variables of control, including CPI of the corresponding consumption category, lagged log value of household expenditure per capita of the consumption category in question, household average years of schooling, household gender ratio, average age squared/100, dependency ratio, , and urban-rural dummies. Robust standard errors clustered at the county level are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1. ··46    To access full collection, visit the World Bank Documents & Report in the Poverty & Equity Global Practice Working Paper series list. www.worldbank.org/poverty