3.6.1 Littafi tambayar
Haxe your binciken to digital burbushi iya zama kamar tambayar kowa da kowa your tambayoyi a duk lokacin da.
Tambayar kullum zo a cikin biyu main Categories: sample safiyo da censuses. Sample safiyo, inda za ka samun damar karamin yawan mutane, zai iya zama m, dace, kuma mun gwada cheap. Duk da haka, sample safiyo, saboda ana dogara ne a kan wani samfurin, sukan iyakance a cikin ƙuduri. tare da samfuri binciken, shi ne sau da yawa wuya a yi kimomi game da takamaiman yanayin yankuna ko takamaiman alƙaluma kungiyoyin. Censuses, a kan wasu, ƙoƙari domin yin tambayoyi kowa da kowa a cikin yawan. Suna da manyan ƙuduri, amma sun kasance kullum tsada, kunkuntar a mayar da hankali (da suka sani kawai sun hada da wani karamin yawan tambayoyi), kuma ba dace (sunã faru a kan gyarawa jadawalin, kamar kowane 10 years) (Kish 1979) . Yanzu tunanin idan masu bincike zai iya hada da mafi kyaun halaye na samfurin safiyo da censuses. tunanin idan masu bincike zai iya tambayar kowane tambaya ga kowa da kowa a kowace rana.
Babu shakka, wannan riƙa, ubiquitous, ko da yaushe-on binciken ne wani nau'i ne na ilmin zaman fantasy. Amma, wannan ya nuna cewa za mu iya fara kimanin wannan ta hada binciken tambayoyi daga wani karamin yawan mutanen da digital burbushi daga mutane da yawa. Ina kira da irin wannan hade Littafi tambayar. Idan aikata da kyau, zai iya taimaka mana mu bayar da kimanta cewa sun fi gida (na karami yanayin yankunan), more granular (ga takamaiman alƙaluma kungiyoyin), kuma mafi dace.
Daya misali Littafi roƙa ya zo daga aikin Joshua Blumenstock, wanda ya so ya tattara bayanai da zai taimaka shiryarwa ci gaba a matalauta kasashen. More musamman, Blumenstock so ya haifar da wani tsarin auna arziki da alheri da hade da cikawa na ƙidaya tare da sassauci da kuma mita wani binciken (Blumenstock 2014; Blumenstock, Cadamuro, and On 2015) . A gaskiya, na riga aka bayyana Blumenstock aikin taƙaice a Babi na 1.
Don fara, Blumenstock gwiwa da most wayar hannu bada a Rwanda. Kamfanin samar da shi anonymized ma'amala records daga game da miliyan 1.5 abokan ciniki rufe hali daga 2005 da kuma 2009. A rajistan ayyukan dauke da bayani game da kowane kira da saƙon rubutu kamar fara lokaci, tsawon lokaci, da kuma m yanayin wurin da mai kira da karɓa. Kafin mu fara magana ne game da ilimin kididdiga al'amurran da suka shafi, yana da daraja nuna cewa wannan mataki na farko zai iya zama daya daga cikin mafi wuya. Kamar yadda aka bayyana a Babi na 2, mafi digital alama data ne m to masu bincike. Kuma, mutane da yawa da kamfanoni ne justifiably komo acikin raba su data domin shi ne masu zaman kansu. da yake su abokan ciniki yiwuwa bai sa ran cewa da records za a rabawa-in girma-da masu bincike. A wannan yanayin, da masu bincike ya mai da hankali matakai don anonymize da bayanai da aikin da aka ke kula da wani ɓangare na uku (ie, su IRB). Amma, duk da irin wadannan kokari, wadannan bayanai ne mai yiwuwa har yanzu tabbatarwa kuma suka m dauke da m bayanai (Mayer, Mutchler, and Mitchell 2016; Landau 2016) . Zan koma can da'a tambaya a Babi na 6.
Ka tuna cewa Blumenstock sha'awar aunawa arziki da alheri. Amma, wadannan halaye ne ba kai tsaye a cikin kira records. A takaice, wadannan kira records ne bai cika wannan bincike, a kowa alama na digital burbushi da aka tattauna a daki-daki, a Babi na 2. Amma, ga alama m cewa kira records yiwuwa da wasu bayanai game da dũkiya da alheri. To, daya hanyar tambayar Blumenstock tambaya zai iya zama: shi ne zai yiwu don hango ko hasashen yadda wani zai amsa a duba bisa laákari da digital alama data? Idan haka ne, to, da tambayar 'yan mutane, za mu iya tsammani amsoshi na kowa da kowa kuma.
Don tantance wannan empirically, Blumenstock da bincike mataimakansa daga Kigali Cibiyar Kimiyya da Fasaha da ake kira samfurin game da dubu wayar hannu abokan ciniki. The masu bincike bayyana a raga na aikin da mahalarta, ya nemi su yarda danganta binciken martani ga kiran records, sa'an nan kuma ya tambaye su a jerin tambayoyi don auna dũkiyõyinsu da alheri, kamar "Kada ka mallaka a rediyo? "da kuma" Shin, ka mallaka keke? "(duba siffa 3.11 domin a m list). All mahalarta a cikin binciken da aka cika financially.
Next, Blumenstock amfani da biyu-mataki hanya kowa a data kimiyya: alama injiniya bi da dubawa koyo. Na farko, a cikin siffa injiniya mataki, domin kowa da kowa cewa da aka yi hira, Blumenstock tuba kira records a cikin wani sa na halaye kamar kowane mutum. data masana kimiyya iya kiran wadannan halaye "fasali" da kuma zamantakewa masana kimiyya zai kira su "canji." Alal misali, ga kowane mutum, Blumenstock lasafta total number of kwana tare da aiki, yawan jinsin mutane da mutum ya kasance a lamba tare, yawan kudi kashe a airtime, da sauransu. Kafofin yada, mai kyau alama aikin injiniya na bukatar sanin bincike saitin. Alal misali, idan yana da muhimmanci rarrabe tsakanin cikin gida da kuma na kasa da kasa da kira (mu iya sa ran mutanen da suka kira duniya su zama arziki), to, wannan dole ne a yi a alama injiniya mataki. A bincike da kadan fahimtar Rwanda zai ba sun hada da wannan alama, sa'an nan kuma gaibu yi na model zai sha wahala.
Next, a cikin dubawa koyo mataki, Blumenstock gina ilimin kididdiga model hango ko hasashen binciken amsa ga kowane mutum bisa laákari da fasali. A wannan yanayin, Blumenstock amfani kayayyaki komawa da baya tare da 10-Musulunci giciye-Ingancin, amma yana iya yi amfani da wani iri-iri da sauran ilimin kididdiga ko na'ura koyo fuskanci.
To, ta yaya da kyau ba shi aiki? Was Blumenstock iya hango ko hasashen amsoshin nazarin tambayoyi kamar "Kada ka mallaka rediyo?" Da kuma "Kada ka mallaka keke?" Ta amfani da fasali samu daga kira records? Sort of. The daidaito daga cikin tsinkaya kasance high ga wasu halaye (Figure 3.11). Amma, shi ne ko da yaushe da muhimmanci a kwatanta wani hadadden Hasashen Hanyar da mai sauki madadin. A wannan yanayin, mai sauki madadin shi ne ya hango ko hasashen cewa kowa da kowa zai ba da ya fi na kowa amsar. Alal misali, 97,3% ruwaito mallakan rediyo don haka idan Blumenstock ya annabta cewa kowa da kowa zai bayar da rahoton mallakan rediyo zai yi wani daidaito na 97,3%, wanda yake shi ne mamaki kama da wasan kwaikwayon da ya fi hadaddun hanya (97.6% daidaito). A takaice, da dukan zato bayanai da kuma yin tallan kayan kawa karu da daidaito na Hasashen daga 97,3% zuwa 97,6%. Duk da haka, ga wasu tambayoyi, kamar "Shin, ka mallaka keke?", Da tsinkaya inganta daga 54,4% zuwa 67,6%. More kullum, Figure 3.12 shows ga wasu halaye Blumenstock bai inganta da yawa bayan kawai yin sauki baseline Hasashen, amma cewa wasu halaye akwai wasu kyautata.
Figure 3.11: Canke daidaito ga ilimin kididdiga model horar da kira records. Results daga Table 2 of Blumenstock (2014) .
Figure 3.12: kwatanta gaibu daidaito ga ilimin kididdiga model horar da kira records to sauki baseline Hasashen. Points suna dan kadan jittered don kauce wa zoba. ga Table 2 of Blumenstock (2014) ga ainihin dabi'u.
A wannan aya ka iya tunanin cewa wadannan sakamakon ne a bit m, amma kawai shekara guda baya, Blumenstock da biyu abokan aiki-Gabriel Cadamuro da Robert On-wallafa wani takarda a Science da ma mafi alhẽri sakamakon (Blumenstock, Cadamuro, and On 2015) . Akwai biyu main fasaha dalilai domin kyautata: 1) da suka kasance sunã more sophisticated hanyoyin (ie, wani sabon tsarin kula da ƙunshi aikin injiniya da kuma more sophisticated na'ura ilmantarwa model) da kuma 2) maimakon yunkurin infer martani ga mutum binciken tambayoyi (misali, "Kada ka mallaka rediyo?"), su yi yunkurin infer a hadedde dũkiya index.
Blumenstock da kuma abokan aiki nuna wasan kwaikwayon da m a cikin hanyoyi biyu. Na farko, su gano cewa, ga mutane a cikin samfurin, za su iya yi wani kyakkyawan aiki mai kyau da tsinkaya da dũkiyõyinsu daga kira records (Figure 3.14). Na biyu, da kuma har abada mafi mahimmanci, Blumenstock da kuma abokan aiki ya nuna cewa su hanya zai iya samar da high quality-kimomi da yanayin rarraba dũkiyõyi a Rwanda. More musamman, da suka kasance sunã da na'ura ilmantarwa model, wanda aka horar da su a kan samfurin game da mutane 1,000, a hango ko hasashen dũkiyar dukan mutane miliyan 1.5 a kira records. Bugu da ari, tare da geospatial data saka a cikin kira data (tuna cewa kiran data hada da wuri daga cikin mafi kusa cell hasumiya ga kowane kira), da masu bincike sun iya kimanta da m wurin zama na kowane mutum. Tunzura nan biyu kimomi tare, da bincike samar da wani kimanta daga cikin yanayin rarraba saye arziki a musamman lafiya sarari granularity. Alal misali, suna iya kimanta da talakawan dũkiyõyi a kowane daga Rwanda ta 2148 sel (karami administrative naúrar a kasar). Wadannan annabta dũkiya dabi'u sun kasance haka granular sun kasance wuya a duba. Saboda haka, da masu bincike aggregated da sakamakon samar da kimomi da talakawan dũkiyar Rwanda ta 30 gundumomi. Wadannan gundumar-matakin kimomi da aka karfi da alaka da kimomi daga zinariya misali gargajiya binciken, Rwandan alƙaluma da Health Survey (Figure 3.14). Ko da yake kimomi daga biyu kafofin kasance m, da kimomi daga Blumenstock da kuma abokan aiki sun kasance game da sau 50 rahusa da kuma sau 10 sauri (lokacin da kudin a auna cikin sharuddan m halin kaka). Wannan ban mamaki karu a cost nufin cewa maimakon ake gudu kowane 'yan shekaru, kamar yadda shi ne misali ga alƙaluma da Health Safiyo-da matasan kananan binciken a hade tare da babban digital alama data za a iya gudu kowane wata.
Figure 3,13: Schematic na Blumenstock, Cadamuro, and On (2015) . Call data daga wayar kamfanin ya tuba zuwa matrix da daya jere ga kowane mutum da kuma daya shafi ga kowane alama (ie, m). Next, da masu bincike ya gina dubawa ilmantarwa model hango ko hasashen binciken martani daga mutum zuwa alama matrix. Sa'an nan kuma, dubawa ilmantarwa model aka yi amfani da zargi da binciken martani ga kowa da kowa. A ainihi, masu bincike sun yi amfani da martani game da dubu mutane su zargi dũkiyar game da daya da mutane miliyan. Har ila yau, da masu bincike kiyasta kimanin wurin zama ga dukan mutane miliyan 1.5 bisa wurare da kira. A lokacin da wadannan biyu kimomi da aka hade-kiyasta dũkiya da kiyasta wurin zama-sakamakon kasance kama kimomi daga alƙaluma da Health Survey, zinariya-misali gargajiya binciken (Figure 3.14).
Figure 3.14: Results daga Blumenstock, Cadamuro, and On (2015) . A cikin mutum-matakin, da masu bincike sun iya yi wani m aiki a tsinkaya wani ta dũkiya daga kiransu records. The kimomi na gundumar-matakin dũkiya-wanda aka dogara ne a kan mutum-matakin kimomi daga dũkiya da wurin zama-sakamakon kasance kama da sakamakon daga alƙaluma da Health Survey, zinariya-misali gargajiya binciken.
A ƙarshe, Blumenstock ta Littafi tambayar m hade binciken data da digital alama data samar da kimomi m da zinariya-misali duba kimomi. Wannan musamman misali kuma bayyana wasu daga cikin cinikayya-offs tsakanin Littafi roƙa da gargajiya duba hanyoyin. Na farko, da Littafi tambayar kimomi kasance mafi dace, ma mai rahusa, kuma mafi granular. Amma, a daya hannun, a wannan lokaci, babu wani karfi da msar tambayar dalilin irin wannan Littafi roƙa. Wannan ne, wannan misali ba ya nuna a lokacin da zai yi aiki, kuma a lõkacin da ta so ba. Bugu da ari, Littafi roƙa m ba yet da kyau hanyoyin da za a quantify rashin tabbas a kusa da kimomi. Duk da haka, Littafi roƙa yana zurfi sadarwa zuwa uku da manyan yankunan a statistics-model na tushen post-stratification (Little 1993) , imputation (Rubin 2004) , da kuma kananan-area hakkin (Rao and Molina 2015) -and don haka sai na sa ran cewa ci gaba so zama m.
Littafi roƙa ya bi wani asali girke-girke da za a iya kera to your musamman halin da ake ciki. Akwai biyu sinadaran da matakai biyu. The biyu sinadaran 1) a digital alama dataset da yake fadi amma bakin ciki (wato, yana da mutane da yawa amma ba da bayanin cewa, kana bukatar game da kowace persons) da kuma 2) a duba cewa shi ne kunkuntar amma m (wato, yana kawai 'yan mutane, amma tana da bayanan da ke bukatar game da wadanda mutane). Sa'an nan kuma, akwai matakai biyu. Na farko, ga mutãne, a biyu data kafofin, gina na'ura ilmantarwa model cewa yana amfani da digital alama data hango ko hasashen binciken amsoshi. Next, amfani da na'ura da ilmantarwa model to zargi da binciken amsoshi na kowa da kowa a cikin digital alama data. Saboda haka, idan akwai wata tambaya cewa kana so ka tambayi don kuri'a na mutane, duba ga digital alama data daga waɗanda mutane da zai iya amfani da su hango ko hasashen da amsar.
Gwada Blumenstock na farko da na biyu ƙoƙari a matsalar kuma nuna wani muhimmin darasi game da miƙa mulki daga biyu zamanin to uku zamanin hanyoyin nazarin bincike: farkon ba karshen. Wannan ne, sau da yawa, na farko m ba zai zama mafi kyau, amma idan masu bincike ci gaba da aiki, abubuwa iya samun mafi alhẽri. More kullum, a lõkacin da kimantawa sabon hanyoyin zamantakewa bincike a cikin dijital shekaru, yana da muhimmanci a yi biyu jinsin kimantawa: 1) yadda da ya aikata wannan aikin a yanzu da kuma 2) yadda da kake tsammanin wannan zai yi aiki a nan gaba kamar yadda data wuri mai faɗi canje-canje da kuma yadda masu bincike sadaukar more hankali ga matsalar. Ko da yake, masu bincike suna horar da su yi ta farko irin kimantawa (yadda mai kyau ne wannan musamman yanki na bincike), na biyu shi ne sau da yawa mafi muhimmanci.
