3.4 Ngubani ukubuza
Ubudala bedijithali benza isampula enokwenzeka ekusebenzeni nzima kwaye kudala amathuba amatsha okungenakwenzeka.
Kwimbali yesampulu, kukho iindlela ezimbini ezikhuphisanayo: iindlela zokusampula iindlela kunye nezindlela zokungabikho ezingenakwenzeka. Nangona zombini iinkqubo zisetyenzisiwe kwiintsuku zokuqala zokusampula, ukutshekishwa kweso sampula kuye kwabalawula, kwaye abaninzi abaphandi bezentlalo bafundiswa ukuba babone ukuthembela okungenakwenzeka. Nangona kunjalo, njengoko ndiza kuchaza ngezantsi, utshintsho oludalwe lixesha ledijithali lithetha ukuba ixesha lokuba abaphandi baphinde bacinge ngeempendulo ezingenakwenzeka. Ngokukodwa, ukutshekishwa kwempembelelo kuye kwaba nzima ukukwenza ngokwenza, kwaye isampampululwazi olungenakunokwenzeka luye luza ngokukhawuleza, oluthile, kwaye lubhetele. Uphando olunokukhawuleza kwaye oluncinci aluphelelanga ngokwabo: banceda amathuba amatsha afana nokuphanda rhoqo kunye neesampula ezinkulu. Ngokomzekelo, ngokusebenzisa iindlela ezingenakwenzeka ukuba i-Co-operative Congressional Election Study (CCES) ikwazi ukuba namaqela angaphezu kwama-10 athatha inxaxheba ngaphezu kwezifundo zangaphambili usebenzisa isampula. Isampulu esona sikhulu senza abaphandi bezopolitiko bafunde ukutshintsha kwendlela yokuziphatha kunye nokuziphatha kuwo onke amanqwanqwa kunye nezimo zentlalo. Ukongezelela, konke oku kufakwe kwinqanaba eliza ngaphandle kokunciphisa umgangatho (Ansolabehere and Rivers 2013) .
Njengamanje, indlela ephezulu yokupasa uphando lwezentlalo inokwenzeka ukuba isampulu . Ngethuba lokusasazwa, onke amalungu ejoliswe kubemi anolwazi olunokwenzeka, olungenalo ilungelo lokusampula, kwaye bonke abantu abathintelayo bayasabela kwisaveyi. Xa le miqathango idibeneyo, iziphumo eziphezulu zeemathematika zinikezela ngokuqinisekileyo malunga nokukwazi umphenyi ukusebenzisa isampuli ukwenza inkcazo malunga nabantu abajoliswe kuyo.
Ehlabathini lenene, nangona kunjalo, iimeko eziphantsi kweziphumo zeemathematika azifane zidibene. Ngokomzekelo, kukho iziphoso zokubhalwa kunye nokungahambisani. Ngenxa yezi ngxaki, abaphandi bavame ukuba basebenzise ukulungiswa kwezibalo ezahlukeneyo ukwenzela ukuba benze i-sample kwi-target target. Ngaloo ndlela, kubalulekile ukuhlukanisa phakathi kweso sampula kwi-theory , eneempawu eziqinileyo zeempendulo, kunye neempendulo zengeniso , ezingenayo iziqinisekiso ezinjalo kwaye kuxhomekeke kwiimeko ezahlukeneyo zokuhlengahlengiso.
Ngokuhamba kwexesha, ulwahlulo phakathi kokusasazwa kwimizekelo yesifundo kunye nokunokwenzeka ukuba isampula ekusebenzeni isanda. Ngokomzekelo, izinga lokungahambisani nexabiso liye landa ngokunyuka, nangona uphando oluphezulu, uphando olubi (umzekelo 3.5) (National Research Council 2013; BD Meyer, Mok, and Sullivan 2015) . Iziphumo zeNonresponse ziphakamileyo kakhulu kwiimvavanyo zethengiselwano zorhwebo-ngamanye amaxesha ziphakamileyo zibe ngama-90% (Kohut et al. 2012) . Ezi zinyuka ngokungaboniyo zisongela umgangatho weengqikelelo kuba uqikelelo luxhomekeka ngokuxhomekeke kwimilinganiselo yamanani abaphandi abasebenzisa ukulungelelanisa ukungabi nantoni. Ukongezelela, oku kunciphisa umgangatho okwenziwe nangona kunzima kakhulu ukunyusa abaphandi abaphandi ukuba balondoloze izinga eliphezulu lokuphendula. Abanye abantu bayesaba ukuba le mihla yokuma kwexabiso elinciphileyo kunye neendleko ezikhulayo zisongela uphando lophando (National Research Council 2013) .
Umzekeliso 3.5: Ukungabi nxamnye kuye kwanda ngokukhawuleza, kwimiba ephezulu ephezulu yokuhlola (National Research Council 2013; BD Meyer, Mok, and Sullivan 2015) . Iziphumo zeNonresponse ziphezulu kakhulu kwiingxelo zeefowuni zorhwebo, ngamanye amaxesha ziphakamileyo njenge 90% (Kohut et al. 2012) . Ezi zihlandlo zangexesha elide ngaphandle kokungathethi kuthetha ukuba ukuqokelela kwedatha kuyabiza kakhulu kwaye ukuqikelelwa kungathembeki kangako. Ukutshintshwa kwi- BD Meyer, Mok, and Sullivan (2015) , umfanekiso 1.
Ngethuba elifanayo lokuba kukho ubunzima obukhulayo beendlela zokunzululwa kweempendulo, kuye kwaba nefuthe elonwabisayo kwiindlela ezingenakwenzeka zokutshekisha . Kukho iintlobo ezahlukeneyo zezindlela zokuncedisa, kodwa enye into abafana nayo kukuba ayinakukwazi ukufikelela ngokufanelekileyo kwisakhelo semathematika sempendulo (Baker et al. 2013) . Ngamanye amazwi, kwiindlela zokungabikho ezingenakwenzeka ukuba wonke umntu unako ukukwazi nokungaxhomeki kwithuba lokufakwa. Izindlela zokungabikho ezingenakwenzeka zineenkcukacha ezibuhlungu phakathi kwabaphandi bezentlalo kwaye zidibaniswa nezinye zezinto ezingabonakaliyo abaphandi abaphandi, njengengxelo ye- Literary Digest (ekuthethwa ngayo ngaphambili) kunye ne "Dewey Intshabalalisa iTruman," ukuchazwa okungafanelekanga nge-US ukhetho lukaMongameli luka-1948 (umzekeliso 3.6).
Umzekeliso 3.6: UMongameli uHarry Truman ephethe isihloko esiphezulu sephephandaba elimemezele ngokungalunganga ukutshatyalaliswa kwakhe. Esi sihloko sisekelwe kwinxalenye ekuqikelelweni kwiisampula ezingenakwenzeka (Mosteller 1949; Bean 1950; Freedman, Pisani, and Purves 2007) . Nangona i-"Dewey Defeats Truman" yenzeke ngowe-1948, kusekhona esinye isizathu sokuba abanye baphandi banokungathembeki malunga nokuqikelelwa kweesampuli ezingenakwenzeka. Umthombo: I- Harry S. Truman Library kunye neMyuziyam .
Olunye uhlobo lwesampampu olungenakunokwenzeka oluthile olufanelekileyo kwixesha ledijithali lusetyenziso lweepaneli ze- intanethi . Abaphandi abasebenzisa iipaneli ze-intanethi baxhomekeke kumniki-mpahla othile-ngokuqhelekileyo inkampani, urhulumente, okanye iyunivesiti-ukwakha iqela elikhulu labantu abahlukeneyo abavuma ukukhonza njengabaphenduli beengxelo. Laba bathathi-nxaxheba bepaneli bavame ukuqashwa ngokusebenzisa iindlela ezahlukeneyo ezifana neentengiso ze-banner. Emva koko, umphandi unakho ukuhlawula umniki-mpahla wokufikelela kwisampuli yabaphenduli abaneempawu ezifunayo (umzekelo, ukumela abantu abadala). La mapaneli e-intanethi ayikho iindlela ezingenakwenzeka ngenxa yokuba akusiwo wonke umntu owaziwayo, ongenalo ilungelo lokufakwa. Nangona ama-panels angabonakaliyo kwi-intanethi sele esetyenziswe ngabaphandi bezentlalo (umz., I-CCES), kusekho ingxoxo malunga nomgangatho weengqikelelo ezivela kubo (Callegaro et al. 2014) .
Naphezu kwezi mpikiswano, ndicinga ukuba zimbini izizathu zokuba ixesha lilungile ukuba baphandi bezentlalo baphinde bacinge ngeempendulo ezingenakwenzeka. Okokuqala, kwi-digital age, kukho intuthuko eninzi kwiqoqo kunye nokuhlalutya kwamasampuli angenakwenzeka. Ezi ndlela zintsha zihluke ngokwaneleyo kwiindlela ezabangela iingxaki kwixesha elidlulileyo ukuba ndicinga ukuba kunengqiqo ukucinga ngabo "ngokungenakunokwenzeka ukutshekisha kweempendulo 2.0." Isizathu sesibini sokuba abaphandi bafanele baqwalasele kwakhona isampula esingenakunokwenzeka kuba sampula ukuziqhelanisa kuya kuba nzima. Xa kukho amanani aphakamileyo okungaphenduli-njengokuba kukho uphando lokwenene ngoku-okubonakalayo kwamathuba okubandakanywa kwabaphenduli awaziwa, kwaye ngoko, amasampula amathandathu kunye neesampula ezingenakwenzeka azinjalo njengabaphandi abaninzi abakholelwa.
Njengoko ndatshilo kwangaphambili, iisampuli ezingenakwenzekayo zijongwa ngokungathandabuzeki ngabaphengululi abaninzi bezenhlalakahle, ngenxa yendima yabo kwezinye zezona zinto zinokuthotyelwa kweentloni kwiintsuku zokuqala zophando. Umzekelo ocacileyo wendlela esikude ngayo nangama-sampula angenakwenzeka kukuba uphando olwenziwa ngu-Wei Wang, uDavid Rothschild, uSharad Goel, no-Andrew Gelman (2015) ngokuchanekileyo bafumene umphumo wokhetho luka-2012 lwase-US basebenzisa isampula esingenako Abasebenzisi be-Xbox baseMelika-isampula esinqumlekileyo esingabonakaliyo baseMerika. Abaphandi bafumana abaphenduli kwi-XBox yokudlala, kwaye njengokuba ungalindela, i-Xbox isampula yindoda kunye nentombazana eneminyaka engama-18: i-19% ye-electorate kodwa i-65% ye-Xbox isampula kunye namadoda yenza i-47% ye-electorate kodwa i-93% yesampula ye-Xbox (umfanekiso 3.7). Ngenxa yale miba ekhuseleyo yabemi, i-Xbox yedatha ebomvu yayingumqondiso ongezantsi wokubuyisela ukhetho. Kwakuxela ukuba u-Mitt Romney onokuqinela ngoBarack Obama. Kwakhona, lo ngowomnye umzekelo weengozi zeesampuli ezingenakwenzeka, ezingenakulungiswa kwaye zikhumbuza i- Literary Digest fiasco.
Umzekeliso 3.7: Amanani abantu abaphenduliweyo W. Wang et al. (2015) . Ngenxa yokuba abaphenduli bebanjwe kwi-XBox, babenokuba bancinane kwaye banokwenzeka ukuba babe ngabesilisa, ngokumalunga nabavoti ngonyulo luka-2012. Utshintshwe W. Wang et al. (2015) , umfanekiso 1.
Nangona kunjalo, uWang kunye noogxa baqaphela ezi ngxaki kwaye bazama ukulungelelanisa inkqubo yabo yokungathathi ngokukhawuleza xa besenza uqikelelo. Ngokukodwa, basebenzise i- post-stratification , isicwangciso esetyenziswa ngokubanzi ukulungelelanisa iisampuli ezinokwenzeka ukuba ziphosakele kunye nezimpendulo.
Ingcamango ephambili ye-post-stratification kukusebenzisa ulwazi oluncedisayo malunga noluntu ekujoliswe kuzo ukulungiselela ukuphucula uqikelelo oluvela kwisampuli. Xa usebenzisa i-post-stratification yokwenza uqikelelo kwi-sampula yabo engenakwenzeka, uWang kunye nomlingani wakhe baqhekeza uluntu kwiqela elihlukeneyo, balinganisela inkxaso ka-Obama kwinqanaba ngalinye, kwaye ke bathatha umyinge olinganisiweyo weengqikelelo zeqela ukuvelisa ukuqikelela. Ngokomzekelo, banokuhlukanisa uluntu ngamacandelo amabini (amadoda nabasetyhini), baqikelelwa ngenkxaso ka-Obama phakathi kwamadoda nabasetyhini, kwaye baqikelelwa ngenkxaso jikelele ka-Obama ngokuthatha isilinganiso esiqingatha ukwenzela ukuba i-akhawunti ibone ukuba abafazi benza i-53% yabavoti kunye namadoda angama-47%. Ngokukhawuleza, ukulungiswa kwe-post-stratification kunceda ukulungelelanisa isampula esingalinganiyo ngokuzisa ulwazi oluncedisayo malunga nobukhulu bamaqela.
Isihluthulelo sokuqhawula i-post-stratification kukukwakha amaqela afanelekileyo. Ukuba unako ukucima uluntu ukuba lube ngamaqela ahambelana nokokuba iimpembelelo zokuphendula zifana nomntu ngamnye kwiqela ngalinye, emva koko i-post-stratification izakuvelisa ukulinganiswa okungenakulungelelaniswa. Ngamanye amazwi, ukulungelelaniswa kwesigxina ngesini kuya kuvelisa ukuqikelelwa okungenakulungelelaniswa ukuba bonke abantu baneempembelelo zokuphendula kunye nabo bonke abafazi baneempendulo ezifanayo. Oku kucatshulwa kuthiwa yi- homogeneous-response-propensities-ngaphakathi kwamacandelo okucinga, kwaye ndiyichaza ngokukodwa kumanqaku eemathematika ekupheleni kwesi sahluko.
Ngokuqinisekileyo, kubonakala ngathi akunakwenzeka ukuba iimpendulo zempendulo ziya kufanayo kubo bonke abantu nabo bonke abafazi. Nangona kunjalo, i-homogeneous-response-propensities-ngaphakathi-amaqela kuthathwa ngokugqithisileyo njengoko inani lamaqela landa. Ngokukhawuleza, kuba lula ukunqumla abantu ukuba babe ngamaqela afanayo xa udala amaqela amaninzi. Ngokomzekelo, kungabonakala kungenakwenzeka ukuba bonke abafazi baneempembelelo ezifanayo, kodwa kunokubonakala kubonakala ukuba kukho ukuphendula okufanayo kubo bonke abafazi abaneminyaka engama-18-29 ubudala, abaphumelele kwiikholeji, nabahlala eCalifornia. . Ngaloo ndlela, njengoko inani lamaqela asetshenziswe kwi-post-stratification liba likhulu, ukucinga okufunekayo ukuxhasa indlela ibe yinto enengqiqo. Ngenxa yoko, abaphengululi bafuna ukudala inani elikhulu lamaqela okwenziwa kwe-post-stratification. Nangona kunjalo, njengoko inani lamaqela landa, abaphandi baphumelela kwingxaki eyahlukileyo: i-sparsity yedatha. Ukuba kukho inani elincinci labantu kwiqela ngalinye, ngoko ke uqikelelo luya kuqinisekiswa ngakumbi, kwaye kwimeko enzima apho kukho iqela elingenabo abaphendulayo, emva koko i-stratification iyaphula ngokupheleleyo.
Kukho iindlela ezimbini eziphuma kulo mbambano ophakathi kwe-plausibility ye-homogeneous-response--------groups-assumption kunye nemfuno yesayizi ekhethekileyo yesampula kwiqela ngalinye. Okokuqala, abaphandi banokuqokelela isampula enkulu, eyahlukeneyo, enceda ukuqinisekisa ubungakanani bezesampula kwiqela ngalinye. Okwesibini, bangasebenzisa imodeli yenkcazo eyinkcazo yokwenza uqikelelo kumaqela. Kwaye, ngamanye amaxesha, ngamanye amaxesha abaphandi baphinda bobabini, njengo-Wang kunye noogxa abasebenza nabo ngokufunda kwabo ukhetho ngokusebenzisa abaphenduli be-Xbox.
Ngenxa yokuba besetyenziselwa iindlela zokungcebeleka ezingabonakaliyo kunye ne-intanethi ezilawulwa yikhomputha (ndiya kuthetha ngakumbi malunga noodliwano-jongene nolondolozo lwekhompyutyana kwicandelo 3.5), uWang kunye nabalingane babenokuqokelela kwedatha engabizi kakhulu, okwabanika ukuba baqokele ulwazi oluvela kuma-345,858 abathathi-nxaxheba abodwa , inani elikhulu ngemigangatho yokuvota kokhetho. Ubunzima besampula obukhulu benza ukuba benze iqela elikhulu lamaqela okupakisha. Nangona i-post-stratification ibandakanya ukuhlutha inani labantu kumaqela ngamaqela, uWang kunye nabalingani bakhe bahlula amaqela ngama-176,256 amaqela achazwe ngobulili (iindidi ezi-2), ubuhlanga (4 iindidi), iminyaka (4 iindidi), imfundo (iindidi ezi-4), urhulumente (Iindidi ezingama-51), i-ID yenkampani (iindidi ezi-3), i-ideology (iindidi ezi-3), kunye nevoti ka-2008 (iindidi ezi-3). Ngamanye amagama, ubukhulu babo besampula, obunikwe amandla ngokuqokelela kwedatha ephantsi, babanceda ukuba benze ukucamngca okuthe xaxa kwinkqubo yabo yokuqikelela.
Ngaphandle kwama-345,858 abathathi-nxaxheba abayingqameneyo, nangona kunjalo, kwakukho amaqela amaninzi, apho ama-Wang kunye nabalingane bakhe babengabikho abaphendulayo. Ngako oko, basebenzise ubuchule obubizwa ngokuba yi- multilevel regress ukuqikelela ukuxhasa kwiqela ngalinye. Okubalulekileyo, ukuqikelela inkxaso ka-Obama ngaphakathi kweqela elithile, ukulungiswa kwamanani amaninzi okudibanisa ulwazi oluvela kumaqela amaninzi amanxulumene. Ngokomzekelo, cinga ukuzama ukuqikelela ukuxhaswa kuka-Obama phakathi kwama-Hispanics ase-Female phakathi kweminyaka eyi-18 ne-29 ubudala, abaqeqeshiweyo kwiikholeji, ababhalise kwi-Democrats, abazithengisa njengama-moderation, kwaye bavotela u-Obama ngo-2008. , iqela elikhethekileyo, kwaye kungenzeka ukuba akukho mntu kwisampuli kunye nale mpawu. Ngoko ke, ukwenza uqikelelo malunga neli qela, ukuguqulwa kwamanani amaninzi kusetyenziswa imodeli yokubala ukuhlanganisa iziqikelelo kunye nabantu abafana namaqela afanayo.
Ngaloo ndlela, uWang kunye noogxa basebenzise indlela eyadibanisa i-multilevel regression kunye ne-post-stratification, ngoko babiza iqhinga labo lokuhlukumeza i-postle-stratification okanye, ngokugqithiseleyo, "Mnu. P. "Xa uWang kunye noogxa basebenzise uMnu P. ukwenza uqikelelo kwi-sampula ye-XBox engenakwenzeka, bavelisa ukuqikelela ngokusondeleyo kwenkxaso jikelele ebanjwe ngu-Obama kunyaka-2012 (umzekeliso 3.8). Enyanisweni uqikelelo lwabo luchanekile ngakumbi kunamaqela eembono zoluntu zoluntu. Ngaloo ndlela, kulo mzekelo, utshintsho lwamanani-ngokukodwa uMnu. P. lubonakala ukwenza umsebenzi omhle olungisa ukuchithwa kweedatha ezingenakwenzeka; ukukhawuleza okwakubonakala ngokucacileyo xa ukhangela uqikelelo oluvela kwi-Xbox yedatha engalungiswanga.
Umzekeliso 3.8: Uqikelelo W. Wang et al. (2015) . Isampula ye-XBox engalungiswanga ivelise izilinganiso ezingachanekanga. Kodwa, i-sampuli ye-XBox enzima ivelise ubungqikelelo obuchanekileyo ngakumbi kunomyinge wokuhlolwa kweefowuni ezinokwenzeka. Utshintshwe W. Wang et al. (2015) , amanani 2 no-3.
Kukho izifundo ezimbini ezibalulekileyo ezivela kuWang kunye noogxa. Okokuqala, iisampula ezingenakulungiswa ezingenakwenzeka zingakhokelela ekuqikeleleni okubi; Esi sifundo abaphandi abaninzi baye bakuva ngaphambili. Isifundo sesibini, nangona kunjalo, ukuba iisampuli ezingenakwenzeka, xa zihlaziywe kakuhle, zinokuvelisa ukuqikelelwa okulungileyo; iisampuli ezingenakwenzekayo azifuneki ngokuzenzekelayo kwinto efana ne- Literary Digest fiasco.
Ukuqhubela phambili, ukuba uzama ukugqiba phakathi kokusebenzisa indlela yokufumana isampula kunye neendlela ezingenakunokwenzeka zokujongana nazo ezijongene nobunzima bokukhetha. Ngamanye amaxesha abaphandi bafuna ukulawula ngokukhawuleza nangokukhawuleza (umz., Sebenzisa rhoqo iindlela zokunzululwa kwimizekelo), kodwa kunzima ukuba unike umgaqo onjalo. Abaphandi bajamelana nokukhetha okunzima phakathi kwendlela yokuthatha isampula ezinokwenzeka-eziya kuba zibiza kakhulu kwaye zide ziphumo ezibonisa ukuba zisebenzise-nezindlela ezingenakwenzeka zokutyelela-ezingabizi kwaye zikhawuleza, kodwa ziyazi kakuhle kwaye zininzi. Enye into ecacileyo, nangona kunjalo, ukuba ukuba unyanzelekile ukuba usebenze ngamasampuli angenakwenzeka okanye imithombo engundoqo engabonakaliyo (cinga kwakhona kwiSahluko 2), ngoko kukho isizathu esinamandla sokukholelwa ukuba uqikelelo olwenziwe ngokusebenzisa ukucwangciswa kweposi Amacandelo anxulumene nawo aya kuba ngcono kunokuba angalungiswanga, uqikelelo oluqingqiweyo.
