3.6.1 amplified nug

Txuas koj daim ntawv ntsuam xyuas rau cov khoom siv yuav ua tau zoo nug txhua leej txhua tus koj cov lus nug rau ntawm txhua lub sij hawm.

Nug feem ntau los nyob rau hauv ob lub ntsiab pawg: cov qauv ntawv tshawb fawb thiab censuses. Qauv ntawv tshawb fawb, uas koj nkag tau ib tug me me muaj pes tsawg tus ntawm cov neeg, yuav ua tau saj zawg zog, raws sij hawm, thiab kuj pheej yig. Txawm li cas los, cov qauv kev tshawb fawb, vim hais tias lawv yog raws li nyob rau hauv ib tug qauv, feem ntau txwv nyob rau hauv lawv cov kev daws teeb meem; nrog ib tug qauv daim ntawv ntsuam xyuas, nws yog feem ntau nyuaj rau ua kev kwv yees txog tej cheebtsam los yog rau tej pej xeem pab pawg. Censuses, nyob rau lwm yam, sim tham nrog txhua leej txhua tus nyob rau hauv cov pejxeem. Lawv muaj zoo kev daws teeb meem, tab sis lawv yog feem ntau kim, nqaim nyob rau hauv kom pom tseeb (lawv tsuas muaj ib tug me me muaj pes tsawg tus ntawm cov lus nug), thiab tsis raws sij hawm (lawv tshwm sim nyob rau ib taag lub sij hawm, xws li txhua txhua 10 xyoo) (Kish 1979) . Tam sim no xav txog tej yam yog tias kev soj ntsuam ntawm yuav muab qhov zoo tshaj plaws yam ntxwv ntawm cov qauv ntawv tshawb fawb thiab censuses; xav txog tej yam yog tias kev soj ntsuam ntawm yuav nug txhua lo lus nug rau txhua leej txhua tus txhua txhua hnub.

Obviously, qhov no continual, ubiquitous, yeej ib txwm-on daim ntawv ntsuam xyuas yog ib yam ntawm cov social science fantasy. Tab sis, nws zoo nkaus li uas peb yuav tau pib rau kwv lam no los ntawm combining daim ntawv ntsuam xyuas cov lus nug los ntawm ib tug me me muaj pes tsawg tus ntawm cov neeg uas muaj cov khoom siv los ntawm ntau tus neeg. Kuv hu no hom ntawm kev ua ke nrog amplified nug. Yog hais tias ua zoo, nws yuav pab tau peb muaj kwv yees uas muaj ntau lub zos (rau me me cheeb tsam), ntau granular (rau tej pej xeem pab pawg), thiab ntau raws sij hawm.

Ib tug piv txwv ntawm amplified thov los ntawm lub chaw ua hauj lwm ntawm Yausua Blumenstock, uas xav sau cov ntaub ntawv uas yuav pab tau phau ntawv txoj kev loj hlob nyob rau hauv teb chaws txom nyem. Dua, Blumenstock xav los tsim ib tug system mus ntsuas wealth thiab zoo uas ua ke rau hauv lub tiav ntawm ib tug census nrog cov yooj thiab zaus ntawm ib tug daim ntawv ntsuam xyuas (Blumenstock 2014; Blumenstock, Cadamuro, and On 2015) . Nyob rau hauv qhov tseeb, kuv twb twb tau piav Blumenstock tus ua hauj lwm luv luv nyob rau hauv Tshooj 1.

Yuav pib, Blumenstock koom tes nrog qhov loj tshaj plaws mobile xov tooj muab kev pab nyob rau hauv Rwanda. Lub tuam txhab muab nws anonymized lw ntaub ntawv los ntawm hais txog 1.5 lab cov neeg muas zaub yuav tsum vov tus cwj pwm los ntawm 2005 thiab 2009. Cov cav muaj cov lus qhia txog txhua hu thiab phau ntawv lus xws li lub sij hawm pib, ntev, thiab approximate chaw nyob ntawm ib tus neeg hu thiab receiver. Ua ntej peb yuav pib sib tham txog cov statistical teeb meem, nws yog tsim nyog taw tawm hais tias qhov no yog thawj kauj ruam yuav tsum yog ib qhov nyuaj tshaj plaws. Raws li tau piav nyob rau hauv Tshooj 2, feem ntau cov cim tseg cov ntaub ntawv yog muab tsis cuag rau kev soj ntsuam. Thiab, tuam txhab uas muag yog justifiably tos kom lawv qhia lawv cov ntaub ntawv vim hais tias nws yog private; uas yog lawv cov neeg muas zaub tej zaum nws tsis xav hais tias lawv cov ntaub ntawv yuav muab qhia tawm-nyob rau hauv tej-nrog soj ntsuam. Nyob rau hauv cov ntaub ntawv no, tus neeg tshawb fawb coj ceev faj cov kauj ruam rau anonymize cov ntaub ntawv thiab lawv ua hauj lwm tau saib xwsle ib tug thib peb-tog (ie, lawv IRB). Tab sis, txawm cov kev siv zog, cov ntaub ntawv no yog tej zaum tseem kheej thiab lawv tej zaum yuav muaj lus rhiab (Mayer, Mutchler, and Mitchell 2016; Landau 2016) . Kuv mam li rov qab mus rau cov koom lo lus nug nyob rau hauv Tshooj 6.

Rov qab hais tias Blumenstock yog txaus siab nyob rau hauv ntsuas wealth thiab zoo. Tab sis, cov sijhawm yog tsis ncaj nraim nyob rau hauv lub cov ntaub ntawv hu. Nyob rau hauv lwm yam lus, cov ntaub ntawv hu tsis tiav rau qhov kev tshawb fawb, ib qho feature ntawm cov ib co kua nplaum uas tau sib tham nyob rau hauv kom meej nyob rau hauv Tshooj 2. Tab sis, no mas, yuav hais tias cov ntaub ntawv hu tej zaum muaj ib co lus qhia txog wealth thiab zoo. Yog li ntawd, ib txoj kev nug Blumenstock lo lus nug yuav yog: yog nws tau mus twv seb yuav ua li cas ib tug neeg yuav teb rau daim ntawv soj ntsuam raws li lawv cov cim tseg cov ntaub ntawv? Yog hais tias yog li ntawd, ces los ntawm nug ib tug ob peb cov neeg peb yuav twv cov lus teb ntawm txhua leej txhua tus lwm tus.

Yuav kom ntsuam xyuas no empirically, Blumenstock thiab kev tshawb fawb pab los ntawm Kigali lub koom haum ntawm Science thiab Technology hu ua ib tug qauv ntawm txog ib txhiab mobile xov tooj cov neeg muas zaub. Cov soj ntsuam ntawm piav qhia cov hom phiaj ntawm peb tes num rau cov neeg koom, nug rau cov kev tso cai rau txuas tus daim ntawv ntsuam xyuas cov lus teb mus rau lub cov ntaub ntawv hu, thiab ces hais kom lawv ib tug series ntawm cov lus nug los ntsuas lawv wealth thiab zoo, xws li "Koj muaj ib tug xov tooj cua? "thiab" koj puas muaj ib tug lub tsheb kauj vab? "(saib daim duab 3.11 rau ib tug ib nrab daim ntawv teev). Tag nrho cov neeg nyob rau hauv lub daim ntawv ntsuam xyuas tau compensated nyiaj txiag.

Tom ntej no, Blumenstock siv ib ob-kauj ruam cov txheej txheem ntau nyob rau hauv cov ntaub ntawv science: feature engineering raws li los ntawm saib xyuas kev kawm ntawv. Ua ntej, nyob rau hauv lub feature engineering kauj ruam, rau txhua leej txhua tus uas tau xam phaj, Blumenstock hloov dua siab tshiab rau hauv lub cov ntaub ntawv hu mus rau hauv ib co ntawm cov yam ntxwv hais txog txhua tus neeg; cov ntaub ntawv zaum yuav hu cov yam ntxwv "nta" thiab social zaum yuav hu lawv "cov tsiaj ntawv." Piv txwv li, rau txhua tus neeg, Blumenstock xam tag nrho cov hnub nrog ua si, cov xov tooj ntawm cov neeg ib tug neeg tau nyob rau hauv kev sib cuag nrog, tus nqi ntawm cov nyiaj siv rau airtime, thiab li ntawm. Mob nyav, zoo feature engineering yuav tsum tau kev paub txog cov kev tshawb fawb chaw. Piv txwv li, yog hais tias nws yog ib qho tseem ceeb kom paub qhov txawv ntawm domestic thiab thoob ntiaj teb (peb yuav cia siab tias cov neeg uas hu chaw yuav wealthier), ces qhov no yuav tsum tau ua nyob rau hauv cov feature engineering kauj ruam. Ib tug neeg soj ntsuam nrog me ntsis kev nkag siab ntawm Rwanda tej zaum yuav tsis muaj xws li no feature, thiab ces tus predictive kev kawm ntawv ntawm tus qauv yuav raug kev txom nyem.

Tom ntej no, nyob rau hauv lub saib xyuas kev kawm kauj ruam, Blumenstock los ua ib tug statistical qauv twv seb rau daim ntawv ntsuam xyuas teb rau txhua tus neeg raws li lawv nta. Nyob rau hauv cov ntaub ntawv no, Blumenstock siv logistic regression nrog 10-fold ntoo khaub lig-validation, tab sis nws yuav tau siv ib tug ntau yam ntawm lwm yam statistical los yog tshuab kev kawm le caag.

Yog li ntawd yuav ua li cas zoo ho nws ua hauj lwm? Yog Blumenstock tau twv seb cov lus teb rau daim ntawv ntsuam xyuas cov lus nug xws li "Koj muaj ib tug xov tooj cua?" Thiab "Koj puas muaj ib tug lub tsheb kauj vab?" Siv nta derived ntawm cov ntaub ntawv hu? Tsi. Cov neeg ntawm lub hau kev yuav los siab rau tej sijhawm (Daim duab 3.11). Tab sis, nws yog ib qho tseem ceeb ib txwm los sib piv ib txoj twv ua ntej txoj kev tiv thaiv ib tug yooj yim lwm. Nyob rau hauv cov ntaub ntawv no, ib tug yooj yim mas yog kwv yees hais tias txhua leej txhua tus yuav muab lub feem ntau lo lus teb. Piv txwv li, 97,3% qhia owning ib tug xov tooj cua li ntawd, yog Blumenstock tau kwv yees hais tias txhua leej txhua tus yuav qhia owning ib tug xov tooj cua nws yuav muaj ib tug neeg ntawm 97,3%, uas yog kuj ceeb tias zoo li qhov kev kawm ntawm nws ntau txoj kev (97.6% tseeb). Nyob rau hauv lwm yam lus, tag nrho cov kev zoo nkauj cov ntaub ntawv thiab modeling nce lub laij qhov twv ua ntej ntawm 97,3% mus rau 97,6%. Txawm li cas los, rau lwm yam lus nug, xws li "Koj muaj ib tug lub tsheb kauj vab?", Txoj hau kev yuav zoo los ntawm 54,4% mus rau 67,6%. Feem ntau, Daim duab 3.12 qhia tau hais tias rau ib co sijhawm Blumenstock tsis txhim kho ntau tshaj cia li ua tej yam yooj yim lub hauv paus twv ua ntej, tab sis hais tias rau lwm qhov zoo muaj ib co txhim kho.

Daim duab 3.11: Predictive raug es thiaj qauv kawm ntaub ntawv hu. Tau los ntawm Rooj 2 ntawm Blumenstock (2014) .

Daim duab 3.12: sib piv ntawm predictive raug es thiaj qauv kawm ntaub ntawv hu rau tej yam yooj yim lub hauv paus twv ua ntej. Cov ntsiab lus yog me ntsis jittered kom tsis txhob sib tshooj; saib rooj 2 ntawm Blumenstock (2014) rau caij nyoog tseem ceeb.

Thaum no tus taw tes tej zaum koj yuav tau xav hais tias cov kev tshwm sim yog ib tug me ntsis disappointing, tab sis cia li ib xyoos tom qab ntawd, Blumenstock thiab ob lug txhawb cov miv-Gabriel Cadamuro thiab Robert On-luam tawm ib daim ntawv nyob rau hauv Science nrog tshuam tau zoo (Blumenstock, Cadamuro, and On 2015) . Muaj ob lub ntsiab kev vim li cas rau kev txhim kho: 1) lawv siv ntau sophisticated txoj kev (ie, ib tug tshiab mus kom ze rau feature engineering thiab ib tug ntau sophisticated tshuab kev kawm qauv) thiab 2) es sim xav hais tias cov lus teb rau ib tug neeg daim ntawv ntsuam xyuas cov lus nug (eg, "koj muaj ib tug xov tooj cua?"), lawv sim xav hais tias ib tug puas wealth index.

Blumenstock thiab lug txhawb cov miv pom qhov kev kawm ntawm lawv mus kom ze nyob rau hauv ob txoj kev. Xub thawj, lawv nrhiav tau tias rau cov neeg nyob rau hauv lawv cov qauv, lawv yuav ua li cas ib tug zoo nkauj txoj hauj lwm zoo cua lawv wealth ntawm cov ntaub ntawv hu (Daim duab 3.14). Ob, thiab puas tau ntau tshaj tseem ceeb, Blumenstock thiab lug txhawb cov miv pom tias lawv txoj kev yuav tsim high-zoo kev kwv yees ntawm lub geographic muab faib rau muaj nyiaj nyob rau hauv Rwanda. Dua, lawv siv lawv lub tshuab kev kawm qauv, uas twb kawm nyob rau hauv lawv cov qauv ntawm txog 1,000 tus neeg, twv seb qhov wealth ntawm tag nrho cov 1.5 lab cov neeg nyob rau hauv lub cov ntaub ntawv hu. Ntxiv mus, nrog rau cov geospatial cov ntaub ntawv kos rau hauv cov hu cov ntaub ntawv (rov qab hais tias tus hu cov ntaub ntawv muaj xws li qhov chaw ntawm lub ze tshaj plaws ntawm tes ntauwd rau txhua hu), cov kev soj ntsuam tau los laij rau tus approximate qhov chaw ntawm qhov chaw nyob ntawm txhua tus neeg. Muab cov ob tug kwv ua ke, cov kev tshawb fawb ua ib tug kwv yees ntawm lub geographic muab faib rau neeg yuav wealth ntawm zoo heev ntsig txog granularity. Piv txwv li, lawv yuav kwv yees qhov nruab nrab wealth nyob rau hauv txhua ntawm Rwanda lub 2148 hlwb (cov tsawg tshaj plaws kev unit nyob rau hauv lub teb chaws). Cov kwv yees wealth qhov tseem ceeb thiaj li granular lawv twb nyuaj mus xyuas. Yog li ntawd, tus neeg tshawb fawb aggregated lawv tau mus rau tsim kev kwv yees ntawm qhov nruab nrab wealth ntawm Rwanda lub 30 ntawv. Cov koog tsev kawm ntawv theem kev kwv yees tau xav txog mus rau lub kev kwv yees los ntawm ib tug kub txheej txheem tsoos daim ntawv ntsuam xyuas, lub Rwandan demographic thiab Health Survey (Daim duab 3.14). Txawm hais tias kev kwv yees los ntawm ob qhov chaw twb zoo sib xws, cov kev kwv yees los ntawm Blumenstock thiab lug txhawb cov miv twb txog 50 lub sij hawm pheej yig dua thiab 10 lub sij hawm sai (thaum them nqi nyob rau hauv kev ntsuas nyob rau hauv cov nqe lus ntawm nce mus nce los nqi). Qhov no txaus ntshai txo nyob rau hauv tus nqi txhais tau tias es tsis raug khiav txhua txhua ob peb lub xyoos-raws li yog txheej txheem rau cov pej xeem thiab Health kuaj-lub hybrid ntawm me me daim ntawv ntsuam xyuas ua ke nrog loj cov cim tseg cov ntaub ntawv yuav tsum khiav txhua lub hlis.

Daim duab 3.13: Schematic ntawm Blumenstock, Cadamuro, and On (2015) . Hu rau cov ntaub ntawv los ntawm cov xov tooj lub tuam txhab twb hloov dua siab tshiab mus rau ib tug matrix nrog ib tug kab rau txhua tus neeg thiab ib kem rau txhua feature (ie, nce mus nce los). Tom ntej no, tus neeg tshawb fawb los ua ib tug saib xyuas kev kawm cov qauv los twv seb daim ntawv ntsuam xyuas cov lus teb los ntawm cov neeg los ntawm feature matrix. Tom qab ntawd, saib xyuas kev kawm cov qauv siv los mus impute lub daim ntawv ntsuam xyuas cov lus teb rau txhua leej txhua tus. Nyob rau hauv essence, tus neeg tshawb fawb siv cov lus teb ntawm txog ib txhiab neeg rau impute qhov wealth ntawm txog ib lab cov neeg. Tsis tas li ntawd, tus neeg tshawb fawb kwv yees kwv yees qhov chaw ntawm qhov chaw nyob rau tag nrho cov 1.5 lab cov neeg raws li nyob rau hauv lub qhov chaw ntawm lawv cov hu. Thaum no ob kwv tau ua ke-lub yees wealth thiab lub kwv yees qhov chaw ntawm qhov chaw nyob-soj ntsuam tau zoo ib yam li kev kwv yees los ntawm cov pej xeem thiab Health Survey, ib tug kub-standard tsoos daim ntawv ntsuam xyuas (Daim duab 3.14).

Daim duab 3.14: tau los ntawm Blumenstock, Cadamuro, and On (2015) . Thaum tus neeg-theem, tus neeg tshawb fawb twb tau mus ua ib tug tsim nyog txoj hauj lwm nyob cua ib tug neeg lub wealth los ntawm lawv cov ntaub ntawv hu. Cov kev kwv yees ntawm koog tsev kawm ntawv theem wealth-uas tau raws li rau ib tug neeg-theem kev kwv yees ntawm wealth thiab qhov chaw ntawm qhov chaw nyob-soj ntsuam tau zoo ib yam li tau los ntawm cov pej xeem thiab Health Survey, ib tug kub-standard tsoos daim ntawv ntsuam xyuas.

Nyob rau hauv xaus, Blumenstock lub amplified nug mus kom ze ua ke daim ntawv ntsuam xyuas cov ntaub ntawv nrog cov cim tseg cov ntaub ntawv los mus tsim kev kwv yees piv nrog kub-standard daim ntawv ntsuam xyuas kev kwv yees. Qhov no xyov piv txwv kuj qhia meej txog ib co ntawm cov lag luam tawm-offs ntawm amplified thov thiab tsoos daim ntawv ntsuam xyuas txoj kev. Ua ntej, lub amplified nug kev kwv yees ntau raws sij hawm, tshuam pheej yig dua, thiab ntau granular. Tab sis, nyob rau hauv lwm cov tes, nyob hauv lub sij hawm no, muaj yog tsis yog ib tug muaj zog theoretical lub hauv paus rau qhov zoo ntawm amplified thov. Uas yog, qhov no yog ib tug piv txwv tsis qhia thaum nws yuav ua hauj lwm thiab thaum nws yuav tsis. Ntxiv mus, lub amplified thov mus kom ze tsis tau muaj txoj kev zoo rau yamntxwv uncertainty ib ncig ntawm nws kev kwv yees. Txawm li cas los, amplified thov muaj sib sib zog nqus sib txuas rau peb loj loj nyob rau hauv statistics qauv-raws li post-stratification (Little 1993) , imputation (Rubin 2004) , thiab me me-cheeb tsam kwv yees (Rao and Molina 2015) -thiab li ntawd kuv thiaj xav hais tias kev kawm yuav yuav sai.

Amplified thov raws li nram no ib tug yooj yim daim ntawv qhia uas yuav tsum tau txiav rau koj cov teeb meem. Muaj ob cov khoom xyaw thiab ob kauj ruam. Cov khoom xyaw ob yog 1) ib cov cim tseg dataset uas yog dav tab sis nyias (uas yog, nws muaj ntau tus neeg tab sis tsis cov ntaub ntawv uas koj xav tau hais txog txhua tus neeg) thiab 2) ib daim ntawv ntsuam uas yog nqaim tab sis tuab (uas yog, nws muaj tsuas muaj ob peb tug neeg, tiam sis nws muaj cov lus qhia uas koj xav tau hais txog cov neeg). Ces, muaj yog ob kauj ruam. Ua ntej, rau cov neeg nyob rau hauv ob qho tag nrho cov ntaub ntawv qhov chaw, tsim ib lub tshuab kev kawm cov qauv uas siv cov cim tseg cov ntaub ntawv twv seb daim ntawv ntsuam xyuas cov lus teb. Tom ntej no, siv tau cov uas tshuab kev kawm qauv rau impute lub daim ntawv ntsuam xyuas cov lus teb ntawm txhua leej txhua tus nyob rau hauv lub digital muaj ib txoj lw cov ntaub ntawv. Yog li, Yog hais tias muaj yog ib co lus nug uas koj xav nug rau ntau ntau tus neeg, saib rau cov cim tseg cov ntaub ntawv los ntawm cov neeg uas tej zaum yuav siv los twv seb lawv teb li.

Muab piv Blumenstock tus thawj thiab thib ob los ntawm qhov teeb meem kuj tso ib zaj lus tseem ceeb txog cov kev hloov los ntawm ob era rau peb era le caag los daim ntawv ntsuam xyuas kev tshawb fawb: pib tsis yog qhov kawg. Uas yog, muaj ntau zaus, tus thawj kauj tsis yuav qhov zoo tshaj plaws, tab sis yog tias kev soj ntsuam ntawm ntxiv ua hauj lwm, tej yam uas yuav tau zoo dua. Feem ntau, thaum ntsuam tshiab le caag los kev tshawb fawb nyob rau hauv lub digital muaj hnub nyoog, nws yog ib qho tseem ceeb los mus ua kom ob cov kev soj ntsuam los: 1) yuav ua li cas zoo li no ua hauj lwm tam sim no thiab 2) yuav ua li cas zoo li koj xav tias qhov no tej zaum yuav ua hauj lwm nyob rau hauv lub neej yav tom ntej raws li cov ntaub ntawv toj roob hauv pes hloov thiab raws li kev soj ntsuam ntawm dedicate rau ntau xim rau qhov teeb meem. Txawm tias, soj ntsuam kev kawm los ua tus thawj hom kev ntsuam (yuav ua li cas zoo yog no xyov daim ntawm kev tshawb fawb), cov thib ob yog feem ntau ntau ib qho tseem ceeb.