3.4.2 Non-Yiwuwar samfurori: weighting

Da wadanda ba yiwuwa samfurori, nauyi iya magance hargitsi sa da zaci daukan samfur tsari.

A wannan hanyar da bincike nauyi martani daga yiwuwa samfurori, su kuma iya nauyi martani daga wadanda ba yiwuwa samfurori. Alal misali, a matsayin madadin zuwa CPS, tunanin cewa ka sanya banner talla a kan dubban yanar kurtu mahalarta ga wani binciken to kimanta da rashin aikin yi kudi. Halitta, za ka zama m cewa sauki nufin ka samfurin zai zama mai kyau kimanta na rashin aikin yi kudi. Your shakka ne mai yiwuwa saboda ka yi tunanin cewa wasu mutane ne mafi kusantar su kammala binciken fiye da wasu. Alal misali, mutane ne da ba su kashe lokaci mai tsawo a kan yanar gizo ne m iya kammala your binciken.

Kamar yadda muka gani a cikin shekaru section, duk da haka, idan muka san yadda samfurin da aka zaba-kamar yadda muka yi da yiwuwar samfurori-to za mu iya magance hargitsi sa da daukan samfur tsari. Abin baƙin ciki, a lõkacin da aiki tare da wadanda ba yiwuwa samfurori, ba mu san yadda samfurin da aka zaba. Amma, za mu iya yi zaton game da daukan samfur tsari, sa'an nan kuma tambaya weighting a cikin wannan hanya. Idan wadannan zaton daidai ne, to, weighting zai kawar da hargitsi ya sa ta daukan samfur tsari.

Alal misali, tunanin cewa a cikin mayar da martani ga m banner ads, ka dauka 100,000 weights. Duk da haka, ba ka yi imani da cewa wadannan 100,000 weights ne mai sauki bazuwar samfurin of American manya. A gaskiya, a lõkacin da ka kwatanta your weights zuwa Amurka yawan, ka samu cewa mutane daga wasu jihohin (misali, New York) ne a kan-wakilta da kuma cewa mutane daga wasu jihohin (misali, Alaska) su ne a karkashin-wakilta. Saboda haka, rashin aikin yi kudi na your sample ne wata ila ya zama wani mummunan kimanta da yawan marasa aikin yi a cikin manufa yawan.

Daya hanyar kawar da murdiya da ya faru a cikin daukan samfur tsari ne sanya kaya masu nauyi zuwa kowane mutum. m nauyi da mutane daga jihohin da aka kan-wakilci a sample (misali, New York), kuma hakan nauyi da mutane daga jihohin da aka karkashin-wakilci a sample (misali, Alaska). More musamman, da nauyi ga kowa wanda ake kara yana da alaka da su ruwan dare a cikin samfurin zumunta ga ruwan dare a Amurka yawan jama'a. Wannan weighting hanya ake kira post-stratification, da kuma ra'ayin yin la'akari kamata tunatar da ku daga cikin misali a Sashe 3.4.1 inda weights daga Rhode Island aka bai kasa nauyi fiye da weights daga California. Post-stratification bukatar ka san isa ya sa ka weights cikin kungiyoyin da sani da rabo daga manufa yawan jama'a a kowane kungiya.

Ko da yake weighting na yiwuwa samfurin da na wadanda ba yiwuwa samfurin iri daya ne da shifran (duba fasaha appendix), suka yi aiki sosai a cikin yanayi daban-daban. Idan bincike yana da cikakken yiwuwa sample (ie, babu ɗaukar hoto kuskure, kuma babu wadanda ba amsa), to, weighting zai samar unbiased kimomi ga dukan dabiu a duk lokuta. Wannan karfi msar tambayar garanti ne dalilin da ya sa masu yada yiwuwa samfurori sãme su sai m. A daya hannun, weighting ba yiwuwa samfurori za su ne kawai nuna unbiased kimomi ga dukan halaye idan amsa propensities ne guda ga kowa da kowa a kowane kungiya. A wasu kalmomin, tunanin mayar da mu misali, ta amfani da post-stratification zai samar unbiased kimomi idan kowa da kowa a New York yana da guda Yiwuwar halartar da kowa da kowa a Alaska yana da guda Yiwuwar halartar da sauransu. Wannan zato ne ake kira kama-amsa-propensities-cikin-kungiyoyin zato, kuma taka wata muhimmiyar rawa a cikin sanin idan post-stratification za ta yi aiki da kyau tare da wadanda ba yiwuwa samfurori.

Abin baƙin ciki, a cikin misali, kama-amsa-propensities-cikin-kungiyoyin zato alama wuya ya zama gaskiya. Wato, ga alama wuya cewa kowa da kowa a Alaska yana da guda Yiwuwar kasancewa a cikin binciken. Amma, akwai uku da muhimmanci da maki a ci gaba a hankali game da post-stratification, duk abin da yi da shi ze more alamar rahama.

Na farko, kama-amsa-propensities-cikin-kungiyoyin zato ya zama mafi plausible matsayin yawan kungiyoyin ƙaruwa. Kuma, masu bincike ba su iyakance ga kungiyoyin kawai bisa guda yanayin girma. Alal misali, za mu iya haifar da kungiyoyin bisa jihar, shekara, jinsi, da kuma matakin ilimi. Da alama more plausible cewa akwai kama amsa propensities cikin rukuni na 18-29, mace, kwalejin digiri zaune a Alaska fiye da a cikin rukuni na dukan mutane da suke zaune a Alaska. Saboda haka, kamar yadda yawan kungiyoyin amfani da post-stratification ƙaruwa, da zaton da ake bukata, don tallafa wa shi zama mafi m. Aka ba da wannan al'amari, ga alama kamar masu bincike da zai so ya haifar da wata babbar dama kungiyoyi domin post-stratification. Amma, kamar yadda yawan kungiyoyin ƙaruwa, masu bincike gudu a cikin wani daban-daban matsala: data sparsity. Idan akwai kawai karamin yawan mutane a kowane kungiya, to, kimomi zai zama mafi bai tabbata, kuma a cikin matsananci hali inda akwai wata ƙungiya da cewa yana da wani weights, to, post-stratification gaba daya karya down. Akwai hanyoyi biyu daga wannan muhimmi tashin hankali tsakanin plausibility na homogeneous- amsa-propensity-cikin-kungiyoyin zato da bukatar m samfurin masu girma dabam a kowane kungiya. Daya m ne su matsa zuwa wani karin sophisticated ilimin kididdiga model for Ana kirga nauyi da sauran shi ne ya tattara ya fi girma, more bambancin samfurin, wanda taimaka tabbatar m samfurin masu girma dabam a kowane kungiya. Kuma, wani lokacin masu bincike yi duka, kamar yadda na ji bayyana a more daki-daki a kasa.

A karo na biyu shawara a lokacin da yin aiki tare da post-stratification daga wadanda ba yiwuwa samfurori shi ne, kama-amsa-propensity-cikin-kungiyoyin zato An riga an akai-akai yi a lokacin da nazarin yiwuwar samfurori. Dalilin da cewa wannan zato ake bukata domin yiwuwar samfurori a yi shi ne cewa yiwuwar samun samfurori da ba amsa, kuma ya fi na kowa hanya domin daidaitawa ga wadanda ba mayar da martani ne post-stratification kamar yadda aka bayyana a sama. Hakika, kawai saboda mutane da yawa masu bincike yin wani zato ba ya nufin cewa ya kamata ka yi da shi ya yi yawa. Amma, shi ba ya nufin cewa a lokacin da gwada wadanda yiwuwa samfurori to yiwuwa samfurori a yi, dole ne mu ci gaba da tuna cewa duka dogara ne a kan zato da karin bayanai don nuna kimomi. A mafi yawan idon basira saituna, akwai kawai ba zato-free tsarin kula da hasashe.

A karshe, idan ka damu da daya kimanta musamman-in mu misali rashin aikin yi rate-to, kana bukatar wani yanayin weaker fiye da kama-amsa-propensity-cikin-kungiyoyin zato. Musamman, ba ka bukatar zuwa ɗauka cewa kowa da kowa yana da guda amsa propensity, ku ne kawai da bukatar ɗauka cewa akwai wata alaka tsakanin amsa propensity da rashin aikin yi rate cikin kowane kungiya. Hakika, ko da wannan weaker yanayin ba zai rike a wasu yanayi. Alal misali, tunanin kimantawa da rabo daga Amirkawa cewa yin sa aiki. Idan mutane suka yi sa kai aiki ne mafi kusantar su yarda da zama a wani binciken, to, masu bincike so tsare kan-kimanta yawan gudanar da aikin sa, ko da sun yi post-stratification sabawa, sakamakon da aka nuna empirically by Abraham, Helms, and Presser (2009) .

Kamar yadda na ce a baya, ba-Yiwuwar samfurori da ake kyan gani, tare da babban shakka daga zamantakewa masana kimiyya, a sashi saboda rawar da wasu daga cikin mafi m kasawa a farkon zamanin binciken bincike. A misali na yadda nisa mun zo da wadanda ba yiwuwa samfurori ne bincike na Wei Wang, David Rothschild, Sharad Goel, kuma Andrew Gelman cewa daidai dawo dasu da sakamako na 2012 Zaben Amurka ta amfani da wadanda ba yiwuwa samfurin of American Xbox users -A decidedly ba bazuwar samfurin Amirkawa (Wang et al. 2015) . The masu bincike dauka weights daga Xbox caca tsarin, da kuma kamar yadda ka iya sa ran, da Xbox sample Ƙirgar namiji da Ƙirgar matasa: 18 - 29 shekara tsufa gyara 19% na za ~ e, amma 65% na Xbox samfurin da maza gyara 47% na za ~ e da kuma 93% na Xbox sample (Figure 3.4). Saboda wadannan karfi alƙaluma biases, da raw Xbox data kasance wani matalauci nuna alama na zaben ya dawo. Yana annabta karfi nasara ga Mitt Romney a kan Barack Obama. Again, wannan shi ne wani misali na hatsarori da raw, unadjusted ba yiwuwa samfurori da kuma shi ne reminiscent na Literary Digest fiasco.

Figure 3.4: YAWAN JAMA'A weights a Wang et al. (2015) . Saboda weights aka dauka daga Xbox, sun kasance mafi kusantar zama matasa da kuma mafi kusantar zama namiji, dangi masu jefa} uri'a, a cikin shekara ta 2012 zaben.

Duk da haka, Wang da abokan aiki su na sane da wadannan matsaloli da kuma} o} arin nauyi da weights gyara ga daukan samfur tsari. Musamman ma, su yi amfani da more sophisticated tsari na post-stratification na gaya muku game da. Yana da daraja koyo a bit more game da m domin shi gina diraya game post-stratification, da kuma musamman version Wang da abokan aiki amfani da shi ne daya daga cikin mafi m hanyoyin weighting ba yiwuwa samfurori.

A cikin sauki misali game da kimantawa rashin aikin yi a Sashe 3.4.1, mun raba yawan cikin kungiyoyin bisa Jihar zama. Da bambanci, Wang da abokan aiki raba yawan shiga cikin 176.256 kungiyoyin tsare by: jinsi (2 Categories), kabila (4 Categories), shekaru (4 Categories), ilimi (4 Categories), Jihar (51 Categories), jam'iyyar ID (3 Categories), akidar (3 Categories) da kuma 2008 kuri'a (3 Categories). Tare da more kungiyoyin, da masu bincike fatan cewa zai zama ƙara kusantar cikin kowane rukuni, mayar da martani propensity ya uncorrelated tare da goyon bayan Obama. Next, maimakon gina mutum-matakin nauyi, kamar yadda muka yi a cikin misali, Wang da abokan aiki ya yi amfani da hadadden tsarin kimanta da rabo mutane a kowane kungiya cewa za su zabi Obama. A karshe, suka hade wadannan rukuni kimomi da goyon baya tare da sani size kowane rukuni, don samar da kimani overall matakin goyon baya. A takaice, suka yankakken sama da yawan zuwa daban-daban kungiyoyin, kiyasta da goyon baya ga Obama a kowane rukuni, sa'an nan kuma ya dauki wani mai nauyi talakawan kungiyar kimomi, don samar da wani overall kimanta.

Saboda haka, babban kalubale a cikin m ne kimanta da goyon baya ga Obama a kowane daga cikin wadannan 176.256 kungiyoyin. Ko da yake su panel hada 345.858 musamman mahalarta, wata babbar dama ta ma'aunansa na zaben polling, akwai mutane da yawa, mutane da yawa kungiyoyi da abin da Wang da abokan aiki da kusan babu weights. Saboda haka, to kimanta da goyon baya a kowane kungiya su yi amfani da dabara da ake kira multilevel komawa da baya tare da post-stratification, wanda masu bincike ƙauna kira Mr. P. gaske, to kimanta da goyon baya ga Obama cikin wani rukuni, Mr. P. wuraren waha bayanai daga mutane da yawa hankali ya shafi kungiyoyin. Alal misali, ka yi la'akari da kalubale na kimantawa da goyon baya ga Obama tsakanin mace, jama'ar {asar Spain, tsakanin 18-29 years old, suke kwalejin digiri, wanda aka yi wa rajista Democrats, suka kai gano yadda moderates, kuma suka zabe Obama a 2008. Wannan ne mai matukar musamman kungiyar, kuma shi ne zai yiwu cewa akwai wanda a sample da wadannan halaye. Saboda haka, don yin kimomi game da wannan kungiya, Mr. P. wuraren waha tare kiyasin daga mutãne, a sosai m kungiyoyin.

Amfani da wannan bincike dabarun, Wang da abokan aiki sun iya amfani da Xbox ba yiwuwa samfurin sosai a hankali kimanta overall goyon bayan cewa Obama samu a cikin shekara ta 2012 zaben (Figure 3.5). A gaskiya su kimomi kasance mafi m fiye da tara jama'a jin ra'ayoyin jama'a. Saboda haka, a cikin wannan harka, weighting-musamman Mr. P.-alama yi aiki mai kyau gyara biases a ba-Yiwuwar data. biases da ke bayyane a lokacin da ka dubi kimomi daga unadjusted Xbox data.

Figure 3.5: An yi} iyasin, daga Wang et al. (2015) . Unadjusted Xbox sample samar m kimomi. Amma, da mai nauyi Xbox sample samar kimomi da suke more m fiye da matsakaita na yiwuwa tushen tarho safiyo.

Akwai biyu main darussa daga nazarin Wang da abokan aiki. Na farko, unadjusted ba yiwuwa samfurori iya haifar da mummunan kimomi. wannan darasi ne cewa mutane da yawa masu bincike sun ji a da. Duk da haka, na biyu shi ne cewa ba yiwuwa samfurori, a lõkacin da mai nauyi da kyau, zai iya zahiri nuna quite m kimomi. A gaskiya ma, su kimomi kasance mafi m fiye da kimomi daga pollster.com, an tari na more gargajiya zaben zaben.

A karshe, akwai muhimmanci gazawar zuwa ga abin da za mu iya koya daga wannan daya musamman binciken. Just domin post-stratification yi aiki da kyau a cikin wannan batu, babu tabbacin cewa za ta yi aiki da kyau a wasu lokuta. A gaskiya ma, zaben ne watakila daya daga cikin mafi sauki saituna saboda pollsters an nazarin zaben domin kusan shekaru 100, akwai na yau da kullum feedback (za mu iya ganin wanda ya lashe zaben), da kuma jam'iyyar ganewa da kuma alƙaluma halaye ne in mun gwada gaibu na zabe. A wannan aya, za mu rasa m ka'idar da empirical kwarewa su san lokacin da weighting sabawa wadanda ba yiwuwa samfurori zai samar da isasshe m kimomi. Wani abu da yake share, duk da haka, shi ne, idan kana tilasta yin aiki tare da wadanda ba yiwuwa samfurori, to, akwai m dalilin yi imani da cewa gyara kimomi zai zama mafi alhẽri daga da ba-gyara kimomi.