Policy Research Working Paper 9198 Using Mobile Phone Data to Reduce Spread of Disease Sveta Milusheva Development Economics Development Impact Evaluation Group March 2020 Policy Research Working Paper 9198 Abstract While human mobility has important benefits for economic infected traveler contributes to 1.7 additional cases reported growth, it can generate negative externalities. This paper in the health facility at the traveler’s destination. This paper studies the effect of mobility on the spread of disease in a develops a simulation-based policy tool that uses mobile low-incidence setting when people do not internalize their phone data to inform strategic targeting of travelers based risks to others. Using malaria as a case study and 15 billion on their origins and destinations. The simulations suggest mobile phone records across nine million SIM cards, this that targeting informed by mobile phone data could reduce paper causally quantifies the relationship between travel the caseload by 50 percent more than current strategies that and the spread of disease. The estimates indicate that an rely only on previous incidence. This paper is a product of the Development Impact Evaluation Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at smilusheva@worldbank.org. The Policy Research 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. Produced by the Research Support Team Using Mobile Phone Data to Reduce Spread of Disease∗ Sveta Milusheva JEL Classication: rSID sISD tIVD sIVD yIS Keywords: re—lthD ˜ig d—t—D epidemi™sD mo˜ilityD pu˜li™ poli™yD mo˜ile phones ∗ s —m gr—teful to endrew posterD tesse ƒh—piroD —nd h—niel fjorkegren for their ™ontinE uous guid—n™e on this proje™tF s would like to th—nk †i™tor eleg—n—D qirij— forkerD h—vid qli™kD worg—n r—rdyD eri—nn— vegoviniD imily ysterD xi™k ‚ukt—non™h—iD fry™e ƒtein˜ergD endrew „—tem —nd ‡illi—m †iolette for v—lu—˜le feed˜—™kF s —ppre™i—te the ™omments from p—rti™ip—nts —t the ‡orld f—nk efghi gonferen™eD the xi…hg gonferen™eD the €op€ov gonferen™eD the €opul—tion re—lth ƒ™ien™es ‡orkshopD the €ee gonferen™e —nd the frown mi™roe™onomi™s lun™h semin—rF „h—nk you to the fill 8 welind— q—tes pound—tion @gr—nt y€€IIIRUWIA —nd the xsgrrh @gr—nt „QP rh UQQVEPWA for (n—n™i—l supportF enonymous mo˜ile phone d—t— h—ve ˜een m—de —v—il—˜le ˜y yr—nge —nd ƒon—tel within the fr—mework of the hRhEƒeneg—l ™h—llengeF s th—nk €hilippe quinot —nd ‰—kou hieyeD €e„r —nd the x—tion—l w—l—ri— gontrol €rogr—m for providing the d—t— on m—l—ri— in™iden™e —nd helpful feed˜—™kF 1 Introduction sn™re—sing domesti™ —nd intern—tion—l mo˜ility h—s m—gni(ed the dev—st—ting ™onseE quen™es of infe™tious dise—sesX more th—n IIDHHH de—ths from i˜ol—D RRHDHHHEIDQHHDHHH ™—ses of ik— infe™tions in IQ ™ountriesD —nd most re™entlyD more th—n IPSDHRV infe™tions —nd RDTIQ de—ths from gy†shEIW1 —™ross IIU ™ountries @ fogo™h et —lF PHITD ‡ry PHPH—D ‡ry PHPH˜AF xeg—tive extern—lities from mo˜ility —re —lso relev—nt for longEst—nding dise—ses th—t we —re —iming to elimin—teF por ex—mpleD †enezuel—D the (rst ™ountry ™erti(ed ˜y the ‡orld re—lth yrg—niz—tion @‡ryA for elimin—ting m—l—ri— in its most popul—ted —re—s in IWTID exE perien™ed — dr—m—ti™ resurgen™e of the dise—se in PHIT in p—rt due to migr—nt workers in the mining region ˜e™oming si™kD tr—veling homeD —nd spre—ding the dise—se to their home vill—ges —nd ™ities @g—sey PHITAF „his p—per uses — ™—se study of m—l—ri— in ƒeneg—l to demonstr—te how to h—rness ˜ig d—t— to ™—us—lly estim—te the size of this extern—lity of movement —nd —pply the results tow—rds more e'e™tive poli™y t—rgetingF „he methods ™—n ˜e —pplied more ˜ro—dly to inform poli™ies rel—ted to mitig—ting spre—d of infe™tious dise—sesF „he e™onomi™s liter—ture identi(es m—l—ri— er—di™—tion —s h—ving import—nt imp—™ts on —dult in™ome —nd ™onsumption @fle—kley PHIHD gutler et —lF PHIHD †enk—t—r—m—ni PHIPAD re—l est—te we—lth @rong PHIIAD longer term he—lth in™luding ™hroni™ dise—se —nd dis—˜ility @rong PHIQAD test s™ores —nd edu™—tion—l —tt—inment @f—rofskyD enekweD —nd gh—se PHISD vu™—s PHIHD †enk—t—r—m—ni PHIPAF2 xinetyEnine ™ountries h—ve ˜een ™erti(ed ˜y the ‡ry —s m—l—ri— freeY howeverD ƒu˜Eƒ—h—r—n efri™—D whi™h —™™ounted for WQ7 of —ll m—l—ri— de—ths in PHIVD h—s only h—d — single su™™essful ™—se of elimin—tion @World Malaria Report PHIWAF ‡hile previous work h—s studied m—l—ri— preventionGtre—tment in shortEterm settingsD fo™using on the pri™ing of m—l—ri— interventions @tessi™— gohen —nd hup—s PHIHD tessi™— gohenD hup—sD —nd ƒ™h—ner PHISD hup—s PHIRD v—xmin—r—y—n et —lF PHIHD „—rozzi et —lF PHIRA —s well —s the —doption of prevent—tive or —ntiEm—l—ri—l tre—tment @edhv—ryu PHIRD epoueyD €i™oneD —nd 1 Data as of March 12, 2020 2 See Currie and Vogl (2013) and Apouey, Picone, and Wilde (2018) for summaries of the literature. I ‡ilde PHIVD erm—nd et —lF PHIUAD rese—r™h h—s not yet ex—mined ˜eh—vior—l f—™tors th—t m—y ™ontri˜ute to the persisten™e —nd spre—d of m—l—ri— in the longEtermF „his p—per uses novel d—t— to ™—us—lly estim—te — ™onstr—ining f—™tor for elimin—tionX the reintrodu™tion of m—l—ri— into elimin—tion zones ˜y popul—tion movementF ‡hile this phenomenon h—s ˜een do™umented in —t le—st TI ™ountries ˜y the epidemiology liter—tureD this h—s not ˜een done in — ™—us—l fr—mework @tustin gohen et —lF PHIPD vu et —lF PHIRAF sn this rese—r™hD s qu—ntify the neg—tive extern—lity of mo˜ility empiri™—lly using — simil—r setting in ƒeneg—lF „his p—per lever—ges poli™y simul—tions ˜—sed on these estim—tes to show how —ggreg—ted ˜ig d—t— on individu—ls9 geolo™—tion ™—n inform more ™ostEe'e™tive t—rgeting str—tegies to redu™e tr—nsmission gener—ted ˜y popul—tion mo˜ilityD whi™h would ˜e — ™omplement—ry ™omponent of — ™—mp—ign to su™™essfully elimin—te m—l—ri—F „he neg—tive extern—lity from tr—vel is gener—ted when people —re un—w—re of their risks to others ˜e™—use they do not know th—t they —re dise—se ve™torsF3 ‰et given the ˜ene(ts of tr—velD —n inform—tion ™—mp—ign is unlikely to ™—use people to intern—lize the extern—lity —nd ™hoose not to tr—vel to prevent infe™ting othersF we—suring the size of the extern—lity —nd identifying those people th—t ™ontri˜ute the most ™—n —llow for t—rgeted poli™ies th—t ™—n help —ddress this m—rket f—ilureF „he m—in ™h—llenge in estim—ting the size of the extern—lity from mo˜ility is th—t while dise—se tr—nsmission m—y respond qui™kly to ™h—nges in migr—tion p—tternsD existing survey d—t— th—t re™ord these p—tterns —re often infrequent or do not h—ve ™over—ge —™ross — ™ountryF4 „hereforeD the only str—tegy —v—il—˜le to poli™ym—kers to —ddress this extern—lity is using in™iden™e in the previous ye—r to identify where —nd who to t—rgetF „his p—per is —˜le to signi(™—ntly improve on this str—tegy ˜y utilizing — new sour™e of d—t— to tr—™k popul—tion 3 As will be described in more detail in the next section, the long incubation period for malaria allows people to travel without knowing they are infected. Additionally, those in high malaria settings typically develop immunity and do not experience malaria symptoms yet can infect mosquitoes when they travel to low malaria settings. 4 In Senegal, the main ocial source of data on population movement is census data that only includes long-term migration statistics every 10 years. There are other surveys that ask about commuting, such as the Household Mobility Survey of Dakar (EMTASUD), but it is only for one point in time, it is focused only on Dakar, and it was done in 2000 and in 2015). P movement for — l—rge num˜er of people ˜etween he—lth f—™ility ™—t™hment —re—s —t the d—ily levelF s lever—ge mo˜ile phone met—d—t— for WFS million ƒsw ™—rds in ƒeneg—l in PHIQ to extr—™t p—tterns of movement ˜etween di'erent —re—s from the —pproxim—te lo™—tions of IS ˜illion ™—lls —nd textsF por e—™h month —nd he—lth f—™ility —re—D s me—sure the num˜er of in™oming tr—velers from other regions weighted ˜y the in™iden™e of m—l—ri— in these regions —nd the length of time spent in the origin —nd destin—tion to ™—l™ul—te 4expe™ted imported m—l—ri— ™—sesF4 s study —n —re— of ƒeneg—l ™lose to elimin—tion to fo™us on reintrodu™tion e'e™tsF s use — p—nel d—t— str—tegy to estim—te the imp—™t of imported in™iden™e on tot—l m—l—ri— in™iden™e in this lowEm—l—ri— setting using — line—r dyn—mi™ p—nelEd—t— model —nd ™ontrolling for time (xed e'e™tsF sf infe™ted tr—velers only le—d to — m—l—ri— ™—se ˜eing dete™ted in the destin—tion r—ther th—n the origin of the tr—velerD ˜ut do not gener—te —ny extern—lity in the form of —ddition—l m—l—ri— ™—sesD then — st—nd—rd model would predi™t for e—™h expe™ted imported ™—se one more —ddition—l ™—se reported in the destin—tionF snste—dD s (nd th—t one —ddition—l expe™ted imported ™—se of m—l—ri— in — low m—l—ri— —re— le—ds to IFU ™—ses of m—l—ri— reportedD indi™—ting —n extern—lity of FU new ™—sesF qiven th—t migr—tion h—s numerous e™onomi™ —nd so™i—l ˜ene(tsD poli™ym—kers f—™e tr—deEo's ˜etween e™onomi™ growth —nd improving pu˜li™ he—lth in designing poli™ies to redu™e tr—velElinked m—l—ri— ™—sesF „his p—per provides — useful fr—mework for str—tegi™ t—rgeting of highErisk popul—tions in lowEin™iden™e —re—s to redu™e neg—tive extern—lities from tr—vel with minim—l interferen™e to tr—vel p—tternsF „here —re two ™—tegories of t—rgeting ™onsideredX @IA t—rgeting highErisk tr—velers entering — low m—l—ri— —re— from — high m—l—ri— —re— —nd @PA t—rgeting —ll tr—velers in only spe™i(™ —re—s of lowEm—l—ri— regions th—t —re likely to re™eive m—ny highErisk tr—velersF5 ‡ithin e—™h typeD s ™omp—re — str—tegy th—t in™orpor—tes d—ily inform—tion on origins —nd destin—tions of tr—velers from mo˜ile phone d—t— with str—tegies th—t only use inform—tion on in™iden™e in the previous ye—r th—t ™ould ˜e implemented ˜y the government in the —˜sen™e of mo˜ile phone d—t—F „he most ™ostEe'e™tive 5 This paper does not focus on the type of targeting, but examples can include information campaigns targeted to travelers via mobile phone and strategically setting up testing sites. Q str—tegy is to use mo˜ile phone d—t— —nd ™om˜ine the two types of t—rgetingF yn —ver—geD given the existing ˜udget —v—il—˜le for this type of —™tivityD the ™ostE e'e™tive str—tegy using mo˜ile phone d—t— performs over SH7 ˜etter ™omp—red to the next ˜est str—tegy th—t only relies on in™iden™e in the previous ye—rF wy empiri™—l design —™™ounts for ™onfounders ™orrel—ted with movementF …sing r—inf—ll proxies —nd month (xed e'e™tsD s ™ontrol for se—sons —nd holid—ysD whi™h drive — l—rge —mount of migr—tion in ƒeneg—lF s —lso test th—t it is not —n uno˜serv—˜le ™orrel—ted with ˜oth migr—tion —nd m—l—ri— driving the resultsD ˜ut inste—d the ™om˜in—tion of movement —nd the m—l—ri— levels —t origins —nd destin—tionsF eddition—l ™he™ks show th—t the rel—tionship ˜etween imported in™iden™e —nd m—l—ri— in™iden™e is not driven ˜y some other rel—tionship ˜etween origins —nd destin—tions —s well —s to ensure th—t the rel—tionship holds only for m—l—ri— —nd not for other he—lth ™onditionsF s —lso test the imp—™t of future imported in™iden™e on m—l—ri— in™iden™e in the ™urrent month —nd (nd no rel—tionshipF „his p—per ˜uilds on previous he—lth liter—ture th—t h—s est—˜lished tr—vel —s — risk f—™tor for ™ontr—™ting m—l—ri—D @wont—lvo —nd ‚eyn—lEuerol PHHUD vyn™h et —lF PHISD ysorioD „oddD —nd fr—dley PHHRD ƒiri et —lF PHIHD vittrell et —lF PHIQAD ˜y estim—ting the size of the ™—us—l imp—™t from —n expe™ted imported ™—seD whi™h m—kes it possi˜le to ™ondu™t poli™y simul—tions —nd ™omp—re di'erent t—rgeting str—tegiesF „—tem et —lF @PHHWA —nd ve wen—™h et —lF @PHIIA use three months of ™ell phone d—t— to estim—te the m—l—ri— import—tion r—te to —nzi˜—r using — st—ti™ model th—t does not —™™ount for se—son—lity due to the limited time fr—me of their mo˜ile phone d—t—F ƒimil—rlyD ‡esolowskiD i—gleD et —lF @PHIPAD inns —nd emu—si @PHIQAD gh—ng et —lF @PHIVA —nd sh—nt—m—l—l— et —lF @PHIVA —mong othersD do not in™orpor—te se—son—lity in in™iden™e —nd fo™us on identifying potenti—l sour™es —nd sinks ˜—sed on tr—vel p—tterns —nd —nnu—l m—l—ri— prev—len™e d—t—F ‰et fu™keeD „—temD —nd tF wet™—lf @PHIUA point out th—t se—son—l v—ri—tion in ˜iologi™—l f—™tors rel—ted to ™lim—te —nd se—son—l popul—tion movements —re import—nt for m—ny infe™tious dise—ses —nd f—iling to —™™ount for se—son—lity ™ould le—d to mis—llo™—tion of resour™esF R ‡hile ‡esolowskiD ir˜—™hEƒ™hoen˜ergD et —lF @PHIUA look —t se—son—lity of movement p—tterns —™ross ueny—D €—kist—n —nd x—mi˜i—D they only ™onne™t this theoreti™—lly to imp—™t on dise—se —nd do not study the rel—tionship with in™iden™e d—t—F €—pers th—t h—ve ™om˜ined se—son—l mo˜ility d—t— from mo˜ile phones with se—son—l dise—se in™iden™e d—t—D su™h —s ‡esolowskiD gF wet™—lfD et —lF @PHISA for ru˜ell— —nd ‡esolowskiD ureshiD et —lF @PHISA for dengueD h—ve not done so in — ™—us—l fr—meworkF „his p—per ™ontri˜utes to the existing work ˜y —iming to me—sure the ™—us—l rel—tionship —nd size of the e'e™t of imported m—l—ri— using — line—r dyn—mi™ p—nelEd—t— model —nd ™ontrolling for time (xed e'e™tsF „hereforeD in —ddition to the two —re—s —lre—dy identi(ed ˜y the e™onomi™s liter—ture —s ne™ess—ry for m—l—ri— redu™tion"pri™ing —nd —doption of prevent—tive —nd tre—tment interventions"this p—per identi(es t—rgeting of higher risk mo˜ile popul—tions —s — thirdF ‡hile this p—per fo™uses on m—l—ri— elimin—tionD it h—s impli™—tions for other dise—ses whose spre—d h—s ˜een —sso™i—ted with tr—vel @edd— PHITD yster PHIPD€rothero IWUUD f—l™—n et —lF PHHWD ƒtu™kler et —lF PHIID „—mD uh—nD —nd vegidoEuigley PHITAF ƒin™e tr—vel p—tterns studied using ™ell phone d—t— ™ould le—d to the tr—nsmission of —ny ™ommuni™—˜le dise—seD if these d—t— —re o˜t—ined for other ™ountries or for di'erent dise—sesD it is possi˜le to repli™—te the —n—lysis using the methods developed in this p—perF s demonstr—te how new sour™es of ˜ig d—t— ™—n ˜e used to me—sure extern—lities —sso™i—ted with tr—vel to develop more e'e™tive t—rgeting str—tegies th—t ™—n ˜e ™om˜ined with pri™ing —nd —doption poli™iesF „his further exp—nds the use of ˜ig d—t— for development in —re—s su™h —s riskEsh—ring @flumensto™kD i—gleD —nd p—f™h—mps PHITAD me—suring poverty @flumensto™kD g—d—muroD —nd yn PHISD flumensto™k PHITA —nd providing ™redit to the poor @fjorkegren —nd qrissen PHIVAF „he p—per ˜egins ˜y providing some ˜—™kground —nd des™ri˜ing the d—t—F st then goes on to model the link ˜etween m—l—ri— —nd popul—tion movement in se™tion QF ƒe™tion R outlines the empiri™—l results linking tr—vel to m—l—ri— —nd se™tion S ex—mines the ™ost e'e™tiveness of di'erent poli™iesF ƒome ro˜ustness ™he™ks —re provided in se™tion TD —nd the p—per ™on™ludes with se™tion UF S 2 Background and Data 2.1 Malaria Characteristics w—l—ri— is —n infe™tious dise—se th—t requires two hosts!hum—ns —nd mosquitoes!in order to spre—dF „he m—l—ri—l ™y™le for P. falciparumD the p—r—site ™—using IHH per™ent of ™—ses in ƒeneg—lD ™—n t—ke sever—l weeks @World Malaria Report PHIRAF efter —n infe™ted individu—l is ˜it ˜y — mosquitoD there is —n in™u˜—tion period l—sting —round W d—ys within the mosquito @uilleenD eF ‚ossD —nd „F ƒmith PHHTAF6 sf the mosquito survives the in™u˜—tion periodD it ™—n ˜ite —nd infe™t — he—lthy individu—lD —fter whi™h there is — se™ond in™u˜—tion period within the hum—n of —round IS d—ys @hF vF ƒmith —nd w™uenzie PHHRD roshen —nd worse PHHRAF ƒymptoms will —ppe—r —t the end of this period —nd the individu—l will ˜e™ome infe™tiousF7 gom˜ining the two in™u˜—tion periodsD — se™ond—ry ™—se will t—ke —round one month to —ppe—r —fter — prim—ry ™—seF8 „his p—per fo™uses on the role of hum—n ˜eh—vior on spre—d of the dise—seF9 „here —re two ™h—nnels through whi™h popul—tion movement ™—n le—d to spre—d of m—l—ri— in lowE m—l—ri— or elimin—tion zonesF „he (rst is residents of these zones who tr—vel to high m—l—ri— —re—s —nd ˜e™ome infe™ted when ˜it ˜y infe™ted mosquitoesF ƒin™e m—l—ri— symptoms do not —ppe—r for —round two weeksD the resident ™—n tr—vel home feeling he—lthyF yn™e —t the home lo™—tionD the person ™—n ˜e™ome symptom—ti™D —s well —s infe™t mosquitoesF „hese infe™ted mosquitoes ™—n infe™t other individu—ls —nd p—ss on the dise—seF „he se™ond ™h—nnel is visitors or migr—nts th—t live in — high m—l—ri— —re— —nd tr—vel to — low m—l—ri— —re—F eg—inD —t the ˜eginning of their tr—velD these individu—ls might not exhi˜it symptomsD ˜ut ™—n still ˜e ™—rriers of the dise—seF„herefore if they —re ˜it ˜y — mosquito in the low m—l—ri— —re—D they ™ould infe™t th—t mosquito —nd it ™ould in turn infe™t other individu—lsF 6 The incubation period can vary, but two dierent sites in Senegal had an average of 9 days. 7 Unlike other malarial parasites, P. falciparum does not have the potential to lie dormant for months. 8 Details on malaria transmission can be found in Doolan, Dobaño, and Baird 2009, D. L. Smith and McKenzie 2004, Killeen, A. Ross, and T. Smith 2006, Wiser 2010. 9 Average radius of travel for the mosquitoes that carry the malaria parasite in Senegal is only 1-2 km; therefore, mosquito movement is not considered (Russell and Santiago 1934, Thomas, Cross, and Bøgh 2013). T 2.2 Health System and Malaria in Senegal ƒeneg—l is geogr—phi™—lly divided into IR he—lth regionsD under whi™h there —re UT he—lth distri™tsF „he m—in point of servi™e for m—l—ri— ™—ses is the he—lth postF „here —re — tot—l of IDPRU he—lth posts in the ™ountry @€xv€D sxpy‚wD vƒr„w PHISAF sn —dditionD there —re rur—l he—lth huts —nd ™ommunity he—lth workers th—t provide ™—re for those living f—r from — he—lth postD —nd report the ™—ses to the ™losest he—lth postF ƒin™e the est—˜lishment of the x—tion—l w—l—ri— gontrol €rogr—m @€xv€A in IWWSD the progr—m h—s ™oordin—ted — v—riety of me—sures —nd poli™ies th—t h—ve led to — redu™tion in de—ths —ttri˜uted to m—l—ri— from IPFWQ per IHHDHHH people in PHHH to VFPT in PHIQ @€xv€D sxpy‚wD vƒr„w PHISAF gurrentlyD the north of the ™ountry h—s very low in™iden™e —nd is —t the level ™onsidered re—dy for elimin—tion ˜y the ‡orld re—lth yrg—niz—tion @I ™—se per IHHHD known —s the preEelimin—tion ph—seAF sn ™ontr—stD the south still h—s — high ™—se lo—dD with some distri™ts —s high —s PUH ™—ses per IHHHF10 „he heterogeneity ™—n ˜e p—rtly —ttri˜uted to environment—l f—™tors ˜e™—use the r—iny se—son is twi™e —s long in the south —s in the northD whi™h —llows for mosquitoes to ˜reed —nd spre—d the dise—se for longerF xeverthelessD the mosquitoes required to spre—d the dise—se —re —lso present in the low m—l—ri— —re—s @xdi—th et —lF PHIPAF qiven the two distin™t zones in the ™ountryD the qovernment of ƒeneg—l strives to ™ontinue redu™ing the ™—se lo—d in the ƒouthD while —iming to elimin—te it ™ompletely from the xorthF es potenti—lly infe™ted individu—ls tr—vel from the ƒouth to the xorthD thoughD they ™—n hinder elimin—tion e'orts in the xorthF 2.3 Population Movement in Senegal ƒeneg—l h—s l—rge )ows of long term —nd perm—nent migr—tionD with PU7 of the popuE l—tion re™orded —s —n intern—l migr—nt in PHHR @€F hF p—llD g—rreteroD —nd wF ‰F ƒ—rr PHIHAF11 e l—rge p—rt of this migr—tion is rur—l to ur˜—n due to irregul—rity of r—inf—ll —nd degr—d—tion 10 The Appendix contains a map of annual malaria incidence by district. 11 This is comparable to the rest of Sub-Saharan Africa, where 50-80 percent of rural households were estimated to have at least one migrant (Deshingkar and Grimm 2005). U of the e™osystems th—t h—ve imp—™ted —gri™ultur—l —™tivity @€F hF p—llD g—rreteroD —nd wF ‰F ƒ—rr PHIHD qoldsmithD qunj—lD —nd xd—rishik—nye PHHRAF sn turnD this longer term migr—tion ™—n le—d to ™ommuting p—tterns —s people return home to visit f—mily —nd friends or re™eive visitors from home @ghoD wyersD —nd veskove™ PHIIAF po™using on migr—nts in h—k—rD eF ƒF p—ll @IWWVA (nds th—t VU7 of m—le —nd VI7 of fem—le migr—nts visited their home —re—sD with the m—jority of visits o™™urring for holid—ysD f—mily ™eremonies —nd religious festiv—lsF het—iled studies of the tol— ethni™ group in sever—l vill—ges (nds th—t ™ir™ul—r migr—tion pl—ys —n import—nt roleD with over VH7 of unm—rried tol— youth tr—veling to the ™ities in y™to˜er —nd then ™oming ˜—™k ˜efore the ri™e h—rvest in tuneEtuly @vin—res PHHQAF fro—der rese—r™h on youth in ƒeneg—l h—s shown th—t more th—n h—lf of the intern—l migr—tion they eng—ge in is tempor—ry —nd rur—l to rur—l or ur˜—n to ur˜—n @rerrer— —nd ƒ—hn PHIQAF eddition—lly there —re still p—stor—l groups th—t tr—vel within — set territory@edri—nsen PHHVAF …nderst—nding the movement p—tterns within ƒeneg—l is import—nt for thinking through potenti—l ™onfounding f—™tors ˜etween movement —nd m—l—ri—F „he m—jority of the liter—ture points to movement triggered ˜y —gri™ultur—l se—sons —s well —s holid—ysF „hese f—™tors —nd their rel—tionship to m—l—ri— in™iden™e will ˜e dis™ussed in the model se™tionF sn ƒeneg—lD P7 of the popul—tion —re intern—tion—l migr—nts while only IFP7 of the popul—tion emigr—ted from ƒeneg—lF po™using on immigr—tion into ƒeneg—l in PHIQD only HFPQ7 of the popul—tion entered the ™ountryF ‡hile the p—per fo™uses on the role of intern—l migr—tionD the potenti—l imp—™t of intern—tion—l migr—tion will ˜e dis™ussedF 2.4 Malaria Data vowEin™iden™e —re—s ™lose to elimin—tion ™—n experien™e the l—rgest extern—lity from popul—tion mo˜ility for three key re—sonsX @IA without these tr—velers the dise—se ™ould ˜e redu™ed to zero —nd require lower government expendituresY @PA in high m—l—ri— —re—sD people h—ve usu—lly ˜uilt up —n immunity to the dise—seY thereforeD — tr—veler entering — high m—l—ri— —re— is less likely to le—d to — new infe™tion even if he or she infe™ts —ddition—l mosquitoes V in the —re—D while in — low m—l—ri— —re— immunity does not existY —nd @QA the infe™tion in — lowEm—l—ri— —re— is likely to ˜e more severe due to the l—™k of exposure to the dise—seF „hereforeD s fo™us on the p—rt of ƒeneg—l dis™ussed e—rlier th—t is —t — preEelimin—tion st—geF ‡ithin this —re—D s fo™us on (ve of the lowest m—l—ri— distri™ts where d—t— —re dis—ggreg—ted —t the he—lth post level —nd —v—il—˜le for every he—lth post in these distri™tsF w—l—ri— d—t— —re not —v—il—˜le —t this high sp—ti—l resolution for —ny of the other low m—l—ri— distri™tsF „he d—t— ™over IIU he—lth postsF „he —ppendix provides — m—p of the (ve he—lth distri™tsD whi™h s su˜divide into —re—s ˜—sed on the lo™—tion of the he—lth posts —nd ™ell phone towersF re—lth posts in ™lose proximity were grouped together forming QT he—lth post ™—t™hment —re—sF s use in™iden™e d—t— ˜—sed on d—t— ™olle™ted from e—™h he—lth post on —ll new ™—ses in the reporting monthF12 „he use of in™iden™e d—t— is one thing th—t sep—r—tes this p—per from some of the previous work th—t relies on endemi™ity d—t—F „he endemi™ity d—t— —re g—thered from p—r—site r—te surveys in whi™h — r—ndom su˜s—mple of the popul—tion is tested for m—l—ri— p—r—sitesF ‡hen the m—l—ri— prev—len™e is very lowD the likelihood of h—ving — positive ™—se ˜e™omes very sm—llF „hereforeD when fo™using on — lowEm—l—ri— setting to underst—nd imp—™t of mo˜ilityD in™iden™e is — more reli—˜le me—sure @eleg—n— et —lF PHIQD tF wF gohen et —lF PHIQAF €xv€9s work h—s led to — system th—t provides high qu—lity d—t— on m—l—ri— in™iden™e —™ross the ™ountryF sn ƒeneg—lD if —n individu—l feels si™kD usu—lly experien™ing — feverD ™hills —nd f—tigueD she will go to the ™losest he—lth post where she will ˜e tested using — r—pid di—gnosti™ test @‚h„A due to her symptomsF sf she tests positiveD she will ˜e provided with medi™—tion for free to tre—t the dise—seF „hereforeD —ll in™iden™e d—t— used in this p—per ™omes from suspe™ted ™—ses th—t h—ve ˜een tested —nd —re positive for m—l—ri— ˜—sed on the testF por the rest of the ™ountryD these in™iden™e d—t— —re —v—il—˜le monthly —t the he—lth 12 The data used to measure malaria incidence comes from the PNLP and PATH, a non-prot organization working with the PNLP to ght malaria in Senegal through its Malaria Control and Elimination Partnership in Africa (MACEPA). W pigure IX ever—ge wonthly re—lth post g—t™hment ere— w—l—ri— sn™iden™e per IHHH —nd ever—ge wonthly ‚—inf—llD t—n PHIQEhe™ PHIS ˜y re—lth histri™t distri™t levelF „hese d—t— —re used to ™l—ssify the risk of tr—velers ˜—sed on their originF wonthly m—l—ri— in™iden™e per IHHH people is —ver—ged —™ross he—lth post ™—t™hment —re—s within distri™ts for three ye—rs in pigure IF histri™ts on —ver—ge h—ve —round HFI ™—ses per IHHH people per monthF „he (gure overl—ys the monthly ™umul—tive r—inf—ll in ™entimeters —ver—ged —™ross he—lth post ™—t™hment —re—sF13 „he ™omp—rison of ™—ses —nd r—inf—ll demonstr—tes strong se—son—lity of m—l—ri— in ƒeneg—l —nd the ™lose rel—tionship ˜etween r—inf—ll —nd m—l—ri—D with the pe—k of ™—ses —nnu—lly o™™urring one to two months —fter the pe—k in r—inf—llF s model this rel—tionship in the —n—lysis sin™e r—inf—ll ™—n ˜e ™orrel—ted with ˜oth m—l—ri— —nd popul—tion movementF „here —re three m—in ™h—llenges th—t —rise with using ™lini™—l d—t—X in™omplete d—t— reportingD presumptive di—gnosis ˜—sed on symptoms r—ther th—n testing —nd nonEutiliz—tion 13 Rainfall data are from the Climate Prediction Center (2016) Rainfall Estimator for Africa. IH of the pu˜li™ he—lth system @eleg—n— et —lF PHIQAF sn the d—t— only IP out of IDRIT he—lth postEmonth o˜serv—tions —re missing in PHIQF sn —dditionD WW7 of suspe™ted ™—ses were tested p—r—sitologi™—lly in the (ve distri™ts —n—lyzedF ƒin™e ˜oth m—l—ri— ™—ses —nd imported ™—ses —re ™—l™ul—ted ˜—sed on ™—se d—t—D —s long —s utiliz—tion is rel—tively uniform —™ross the ™ountryD it should not ˜i—s resultsF f—sed on the hrƒ d—t— for —ll the regionsD — he—lth f—™ility w—s visited for fever in ™hildren under —ge S in RT7 of ™—ses @exƒh —nd sgp sntern—tion—l PHISAF „he st—nd—rd devi—tion of this utiliz—tion —™ross regions is TFS per™ent—ge pointsF ‡hile in the m—in —n—lysisD s —ssume uniform utiliz—tionD s in™lude — ro˜ustness ™he™k where ™—ses —re s™—led ˜y region—l utiliz—tion in the hrƒF 2.5 Population Movement Data „he d—t— used to me—sure short term movement ™ome from phone re™ords m—de —v—ilE —˜le ˜y ƒon—tel —nd yr—nge in the ™ontext of the h—t— for hevelopment gh—llenge @wontjoye et —lF PHIRAF „he d—t— ™ome from the se™ond ph—se of the gh—llenge —nd ™onsist of IS ˜illion ™—ll —nd text re™ords for ƒeneg—l ˜etween t—nu—ry ID PHIQ —nd he™em˜er QID PHIQ for —ll of ƒon—tel9s user ˜—seF14 „he d—t— ™ont—in inform—tion on —ll ™—lls —nd texts m—de or re™eived ˜y — ƒsw ™—rdD their timeD d—te —nd lo™—tion of the ™losest ™ell phone towerD whi™h en—˜les tr—™king of ƒsws in sp—™e —s they m—ke ™—lls from di'erent tower lo™—tionsF „he d—t— —re —nonymizedD with — r—ndom sh provided th—t m—kes it possi˜le to tr—™k the s—me ƒsw over timeD ˜ut no identifying inform—tion on the individu—lsF yn —ver—ge there —re IDTSU ™—lls or texts per sh during the ye—rD —nd on —ver—ge —n sh h—s — ™—ll or text on ISS d—ysF i—™h tower is —ssigned — he—lth distri™t ˜—sed on its q€ƒ ™oordin—tesF s follow previous liter—ture to —ssign individu—ls — d—ily he—lth distri™t lo™—tion ˜—sed on the ™ell tower of the l—st ™—ll or text of the d—y @‚ukt—non™h—i et —lF PHITAF sn inst—n™es where there —re d—ys with no ™—llsD s repli™—te ‡esolowskiD i—gleD et —lF @PHIPA —nd —ssign the he—lth distri™t lo™—tion of the d—y ™losest to the one missingF15 e he—lth distri™t lo™—tion is —ssigned to e—™h ƒsw 14 At this time it was not possible to obtain more recent data or data from other providers. 15 The appendix includes a robustness check where observations with more than 14 days in a row missing II for every d—y of the ye—rF wovement is de(ned —s — ™h—nge in lo™—tion from one he—lth distri™t to —nother ˜etween two ™onse™utive d—ysF „he popul—tion is highly mo˜ileD with over VH7 of —ll ƒon—tel ƒsw ™—rds t—king —t le—st one trip —nd over — qu—rter million tr—veling on —ver—ge on —ny given d—yF yn —ver—ge —nnu—lly per ƒsw there —re IH di'erent trips to —lmost (ve di'erent he—lth distri™tsF sn —ddition to —ssigning the towers within the study —re— to — he—lth distri™tD s —ssign them to — he—lth post ™—t™hment —re— ˜—sed on their q€ƒ lo™—tionF „hereforeD e—™h tr—veler entering one of the (ve he—lth distri™ts is —ssigned — spe™i(™ he—lth post ™—t™hment —re— ˜—sed on the l—st ™—ll or text of the d—yF16 €—nel — of pigure P shows the —ver—ge num˜er of people entering — he—lth post ™—t™hE ment —re— e—™h d—y —s — per™ent of the popul—tion in th—t ™—t™hment —re— —ver—ged —™ross —ll —re—sD —long with verti™—l lines m—rking sever—l religious holid—ys —nd import—nt pilgrim—gesF „he movement p—tterns l—rgely —lign with the holid—ys —nd pilgrim—gesD whi™h supports the (ndings in eF ƒF p—ll @IWWVA th—t the m—jority of migr—nts to h—k—r visit their home —re— prim—rily for holid—ysD religious festiv—ls —nd f—mily ™eremoniesF yn —ver—ge for —ll the he—lth post ™—t™hment —re—sD —round Q per™ent of the popul—tion of th—t —re— enters on —ny given d—yF „he v—ri—tion in per™ent of people entering ™—n v—ry widely ˜y he—lth post ™—t™hment —re— —nd d—te within — distri™t @€—nel fAF por he—lth post ™—t™hment —re—s where —n imporE t—nt religious le—der residesD on ™ert—in religious holid—ys the num˜er of people entering is ™lose to or over SH7 of the popul—tion of the —re—F por other he—lth postsD the ˜eginning of ™ert—in —gri™ultur—l se—sons or other holid—ys le—d to l—rge jumps in people enteringF „his v—ri—tion m—kes it possi˜le to study the imp—™t of people entering on m—l—ri— ™—ses in these —re—s th—t —re otherwise geogr—phi™—lly ™lose together —nd very simil—rF sn PHIQD ƒon—tel h—d slightly over WFS million unique phone num˜ers on its network while the popul—tion of ƒeneg—l w—s IQFS millionF„here —re two sets of people th—t —re potenti—lly ex™luded from the d—t— —nd need to ˜e —™™ounted for!those without — phone —nd those with are removed. 16 Movements within a district between health post catchment areas are not counted. IP pigure PX €eople intering — re—lth €ost g—t™hment ere— —s — €er™ent of the €opul—tion in the ere— (a) Avg Across All Health Post Catchment Areas (b) Health post Catchment Areas in Richard Toll Notes: Red dash lines in panel (a) represent some important religious holidays and pilgrimages. The scale in panel (b) is bounded at 20%, but for Diama Savoigne and Dabi Tiguette Djoudj, the value in January goes up to around 50%. IQ — phone ˜ut using — di'erent mo˜ile providerF f—sed on the vistening to ƒeneg—l ƒurvey @vƒƒA done in PHIRD IPFQ7 of —dults —ge IV —nd over never use — mo˜ile phoneF ƒon—tel is one of three mo˜ile phone providersF f—sed on the vƒƒD ƒon—tel is the m—in provider for VH7 of those surveyed with — ™ell phoneD —nd VV7 of those with — ™ell phone h—ve — ƒon—tel ƒsw ™—rd @egen™e x—tion—le de l— ƒt—tistique et de l— hémogr—phie E winistére de l9i™onomie PHIRAF „hereforeD only —round IP7 of —dults with — ƒsw —re ex™ludedF gom˜ining the two types of missing —dultsD UUFP7 of —dults —re represented in these d—t—F s ™ondu™t ™he™ks to see how represent—tive the d—t— —re for the two types of users th—t —re not in™luded"those without — ƒsw —nd those with — ƒsw from — di'erent providerF s use the hrƒ survey to ™omp—re mo˜ility p—tterns ˜etween women with —nd without — ™ell phone in the householdF17 „here is no st—tisti™—l di'eren™e in whether — trip longer th—n — month w—s t—ken ˜etween those with —nd without — ™ell phoneF „he women with — ™ell phone in the household h—ve only — slightly higher —ver—ge num˜er of trips t—ken in the l—st ye—rF …sing the vƒƒD s ™omp—re sever—l indi™—tors ˜etween people th—t h—ve — ƒon—tel ƒsw ™—rd —nd those th—t only h—ve — ƒsw ™—rd from —nother providerF f—sed on tEtestsD there is no signi(™—nt di'eren™e ˜etween these ™—tegories of individu—ls ˜—sed on type of prim—ry —™tivity @pEv—lueaHFQRRAD se™tor of prim—ry —™tivity @pEv—lueaHFVTQAD sour™e of drinking w—ter @pEv—lueaHFVTVAD nonEfood expenditures over p—st month@pEv—lueaHFIWQAD —nd nonEfood expenditures over p—st IP months@pEv—lueaHFIISAF „he third missing group is ™hildrenF sf ™hildren tr—velD they —re likely tr—veling with —dults @though they might ˜e tr—veling less on —ver—ge if they do not —lw—ys tr—vel every time —n —dult tr—velsAF ƒin™e there is no d—t— ™omp—ring the short term movement of —dults —nd ™hildren in ƒeneg—lD s use —n —™ross the ˜o—rd weight of IFR to represent the full popul—tion —nd to get —n upper ˜ound on movementF „he weighted d—t— overestim—tes tot—l movement —nd underestim—tes the imp—™t of e—™h tripF …nweighted results —re provided in the ro˜ustness se™tion —s —n upper ˜ound on the e'e™t sizeF 17 The survey does not contain data on men. IR 3 Empirical Model „he empiri™—l spe™i(™—tion is derived from — model of m—l—ri— th—t is ˜—sed on previous models used in hF vF ƒmith —nd w™uenzie @PHHRAD gosner et —lF @PHHWA —nd „orresEƒor—ndo —nd ‚odriguez @IWWUAF pour key —ssumptions —llow me to simplify the model so th—t in™iden™e in the ™urrent month is dependent on in™iden™e in the l—st month using — line—r fun™tion—l formF ixpe™ted imported ™—ses enter —s — line—r —dditive term —s in „orresEƒor—ndo —nd ‚odriguez @IWWUAF „he modelD —ssumptions —nd impli™—tions —re des™ri˜ed in det—il in epE pendix eF s st—rt out estim—ting equ—tion I using yvƒD with imported in™iden™e ™—l™ul—ted using equ—tion PX xit = β1 xit−1 + β2 E(Iit ) + αZit + γi + δt + it @IA 1 E(Iit ) = Tip (xjt Tjp ) @PA Hit j =i pt ∈j sn this modelD xit represents the in™iden™eD or num˜er of hum—ns infe™ted in lo™—tion i —t time t per IHHH people in lo™—tion iF18 vo™—tion i is one of the QT he—lth post ™—t™hment —re—s —nd t is —t the monthly levelF „he β1 p—r—meter estim—ted tells us on —ver—ge how m—ny new ™—ses per IHHH —re gener—ted in the following month from ™—ses in the ™urrent monthF sn ™ontr—st to epidemiologi™—l modelsD where this p—r—meter would ˜e estim—ted sep—r—tely for e—™h —re—D s w—nt to ™—us—lly estim—te the se™ond—ry ™—ses gener—tedF „hereforeD s estim—te —n —ver—ge e'e™t —™ross lo™—tions —nd time in order to ˜e —˜le to in™lude he—lth post —re— (xed e'e™tsD γi —nd month (xed e'e™tsD δt F „his —llows me to ™ontrol for m—l—ri— se—son—lity —nd uno˜serv—˜le ™h—r—™teristi™s of — he—lth post —re— th—t might imp—™t in™iden™eF s use the mo˜ile phone d—t— to ™—l™ul—te expe™ted imported m—l—ri— ™—ses entering —nd divide them ˜y the popul—tion of the —re— they enterD Hit D to ™—l™ul—te the expe™ted imported in™iden™eD E(Iit )F „he likelihood of —n infe™ted ™—se entering depends on the origin of —n 18 Populationis based on the known population of each health facility catchment area. The appendix includes a robustness check where the annual population is adjusted monthly using the number of people entering and leaving each month based on the mobile phone data. IS individu—lD how long the person spent there —nd the length of time in the destin—tionF s m—ke two —ssertionsX IF „he likelihood — personD pt D is infe™ted is ˜—sed on the fr—™tion of the month spent in the origin distri™t jD Tjp D —nd the monthly in™iden™e r—te in jD xjt PF „he ™ontri˜ution of —n imported ™—se to — new lo™—tion is ™—l™ul—ted —s — fr—™tion of time spent in the destin—tion lo™—tion iD Tip „he ™ontri˜ution of person pt D who enters i from j —t time tD to imported ™—ses is ™—l™ul—ted ˜—sed on the in™iden™e of the origin distri™t —nd the length of time spent in origin distri™t j —nd destin—tion he—lth post ™—t™hment —re— iF ynly up to IS d—ys in the origin —nd up to IS d—ys in the destin—tion —re ™onsidered sin™e the hum—n in™u˜—tion period is IS d—ysF „hereforeD Tip is the proportion of IS d—ys pt spends in i —fter entering i —nd Tjp is the proportion of the month up to IS d—ys th—t pt spent in j with monthly m—l—ri— in™iden™e xjt in month tF19 hue to the ™ompli™—ted n—ture of the imported in™iden™e v—ri—˜leD s ˜re—k down wh—t expl—ins the v—ri—tion in this v—ri—˜leF smported in™iden™e ™om˜ines inform—tion on tr—velersD the in™iden™e where they —re ™oming fromD the timing in the destin—tion —nd the originD —nd the popul—tion in the destin—tionF f—sed on — p—rti—l R2 of FRR the month expl—ins — l—rge portion of the v—ri—tionD while FQQ of the v—ri—tion is expl—ined ˜y the he—lth post ™—t™hment —re—F tointlyD they expl—in —˜out h—lf of the v—ri—tion @R2 of HFSTAD implying th—t the other h—lf of the v—ri—tion in the v—ri—˜le of interest is ™oming from — ™om˜in—tion of the month —nd the lo™—tion re™eiving imported ™—sesF „his shows how the identi(™—tion ™omes from the unique ™om˜in—tion of det—iled d—t— on tr—velers —nd the in™iden™e in the originF „he m—trix Zit in™ludes zeroD one —nd two l—gs of r—inf—llD whi™h ™—pture ˜oth the —gri™ultur—l se—sons th—t ™ould in)uen™e movement —nd ™h—nges in m—l—ri— in™iden™e due to 19 Iuse the detailed knowledge of the timing from the mobile phone data to factor in how many of the 15 days were in month t and how many in month t − 1 and use the incidence both in month xjt and xjt−1 to determine the probability the person is infected. IT environment—l f—™torsF eddition—l fun™tion—l forms of r—inf—ll were —lso tested ˜ut did not signi(™—ntly ™h—nge the —n—lysisY thereforeD — line—r fun™tion—l form w—s used for r—inf—llF20 it represents idiosyn™r—ti™ sho™ksF s ™luster errors —t the he—lth post ™—t™hment —re— level to —™™ount for the f—™t th—t errors —re ™orrel—ted within p—nelsF21 „he m—in ™oe0™ients of interest —re β1 —nd β2 D whi™h represent the num˜er of se™ond—ry ™—ses gener—ted ˜y infe™ted tr—velers —nd the num˜er of prim—ry m—l—ri— ™—ses imported ˜y infe™ted tr—velersF 3.1 Identication por my identi(™—tion to ˜e ™orre™tD it is ne™ess—ry th—t within — he—lth post ™—t™hment —re— over timeD —ny idiosyn™r—ti™ sho™ks in m—l—ri— in™iden™e —re not ™orrel—ted with expe™ted imported m—l—ri— in™iden™eF egri™ultur—l se—sons —nd holid—ys —re the two m—jor re—sons for tr—velF egri™ultur—l se—sons —re ™orrel—ted with r—inf—llD —nd —ddition—llyD r—inf—ll ™ould —'e™t the ™onditions for tr—vel @qu—lity of ro—dsAF s ™ontrol for this potenti—l ™onfounder ˜y in™luding r—inf—ll ™ov—ri—tes in my spe™i(™—tionF st is —lso possi˜le th—t holid—ysD whi™h in™re—se popul—tion movementD ™ould —'e™t m—l—ri—F €eople might spend more time outside during the holid—ys —nd ˜e exposed to mosquitoesF s —ddress this potenti—l thre—t to identi(™—tion using — pl—™e˜o test where s s™—le tr—velers ˜y —ver—ge monthly in™iden™e in the ™ountry r—ther th—n ˜y the in™iden™e of their originF s —lso ex—mine the rel—tionship ˜etween p—st —nd future imported ™—ses —nd ™urrent m—l—ri—F pin—llyD the dyn—mi™ p—nel model with — rel—tively short p—nel of IP time periods ™ould introdu™e — ˜i—s if the error term is me™h—ni™—lly ™orrel—ted with the l—gged dependent v—ri—˜le on the right h—nd side @xi™kell IWVIAF s study this ˜y ™omp—ring the (xed e'e™ts model with — r—ndom e'e™ts modelF qiven the results of this ™omp—risonD the preferred spe™i(™—tion used is —n —ugmented version of the erell—noEfond qener—lized wethod of woments estim—tor designed to —ddress situ—tions with sm—ll „D l—rge x p—nelsF 20 The appendix includes results with these dierent specications. 21 I include a robustness check with spatial and panel autocorrelated standard errors. IU 4 Results 4.1 Quantifying the Eect of Imported Cases i—™h imported ™—se of m—l—ri— is —sso™i—ted with IFPQ ™—ses of m—l—ri— in the ™urrent period —nd HFQQH ™—ses in the next period ˜—sed on the (xed e'e™ts model @golumn I of „—˜le IAF „his spe™i(™—tion —ssumes the extern—lity from lo™—lly gener—ted —nd imported ™—ses will ˜e the s—meF s expli™itly test this ˜y in™luding l—gged imported in™iden™e —long with l—gged nonEimported in™iden™e @golumn PAF „he ™oe0™ient on l—gged imported in™iden™e is not signi(™—ntly di'erent from the ™oe0™ient on l—gged lo™—l in™iden™eD whi™h implies th—t there is no di'erenti—l e'e™t ˜etween l—gged imported —nd l—gged lo™—l in™iden™eF s estim—te — r—ndom e'e™ts model to test if there ™ould ˜e — dyn—mi™ p—nel ˜i—s due to the in™lusion of (xed e'e™ts with — rel—tively short p—nel @golumn Q of „—˜le IAF „he ™oe0™ient on imported in™iden™e is sm—llerD while the ™oe0™ient on l—gged in™iden™e is l—rgerF sn using r—ndom e'e™tsD thoughD s —m no longer ™ontrolling for timeEinv—ri—nt ™h—r—™teristi™s of the he—lth post —re—s th—t ™ould ˜e ™orrel—ted with ˜oth imported in™iden™e —nd m—l—ri— in™iden™eF s in™lude sever—l ™h—r—™teristi™s of the he—lth f—™ility —re—sD in™luding popul—tion densityD — dummy for ur˜—n —re—sD —nd — dummy for he—lth f—™ility —re—s th—t —re not —long the ˜order of the ™ountry @golumn RAF sn™luding these ™ov—ri—tesD the ™oe0™ient on imported in™iden™e is ˜igger —nd ™loser to the ™oe0™ient from the (xed e'e™ts modelF e r—usm—n test ™omp—ring the two models (nds they —re signi(™—ntly di'erentF qiven e—™h model h—s potenti—l to ˜e ˜i—sedD sin™e the (xed e'e™ts model might h—ve some dyn—mi™ p—nel ˜i—s while the r—ndom e'e™ts model might h—ve omitted v—ri—˜le ˜i—sD s use —n erell—noE fond spe™i(™—tion @golumn S of „—˜le IAF f—sed on this modelD for e—™h imported ™—se of m—l—ri— per IHHHD there —re IFHW ™—ses per IHHH reportedF sn —dditionD for e—™h l—gged ™—se per IHHHD there is —n —ddition—l HFSTQ of — ™—se gener—ted the following monthF „his —lso represents the neg—tive extern—lity of —n imported ™—se the previous monthF „he epidemiologi™—l model th—t the empiri™—l spe™i(™—tion is ˜—sed on le—ds to sever—l IV „—˜le IX i'e™t of smported w—l—ri— sn™iden™e @IA @PA @QA @RA @SA pixed pixed ‚—ndom ‚—ndom erell—no i'e™ts i'e™ts i'e™ts i'e™ts fond smported sn™iden™e IFPQHBB IFIUSBB HFUWQBB HFVTRBB IFHWRBBB @HFRSPA @HFRUTA @HFQWVA @HFQVIA @HFQSUA v—g sn™iden™e HFQQHBBB HFRQRBBB HFRIVBBB HFSTQBBB @HFHSHWA @HFHSUWA @HFHSSPA @HFIPWA v—g smported sn™iden™e HFRRWB @HFPQIA v—g xonEimported sn™iden™e HFQRHBBB @HFHSUIA ‚—in in ™m HFHHTQW HFHHSWP HFHHQRV HFHHRRW EHFHHHTSR @HFHHWQRA @HFHIHPA @HFHHWHRA @HFHHVSQA @HFHHUWHA v—g ‚—in in ™m HFHPWP HFHPWR HFHPWRB HFHQHHB HFHPTW @HFHIVWA @HFHIVVA @HFHIUUA @HFHIUUA @HFHIVIA v—g P ‚—in in ™m HFHQVIBB HFHQUPBB HFHQUTBBB HFHQVPBBB HFHQIWBB @HFHISVA @HFHISSA @HFHIQVA @HFHIRIA @HFHISTA gonst—nt EHFHQVI EHFHQWT EHFHRVUBB EHFHTPIBB EHFHTVVBB @HFHPWRA @HFHQIUA @HFHPQUA @HFHPTRA @HFHPWHA wonth pi ‰es ‰es ‰es ‰es ‰es re—lth €ost ere— gontrols xo xo xo ‰es xo re—lth €ost x wonth y˜s RQP QWT RQP RQP RQP ‚Esqu—red HFSHW HFSIP r—usm—n „est gomp—ring golumn I —nd golumn R pEv—lueaHFHHVQ „esting €redi™tions „est v—g smported av—g xonEsmported pEv—lueaHFTSH „est smportedaI pEv—lueaHFUWP „est v—g sn™iden™eaHFPUQ pEv—lueaHFHPRU „est smported C v—g sn™iden™eaI pEv—lueaHFHRPV ‚o˜ust st—nd—rd errors in p—rentheses BBB p