2.4.2 kiyasin da nowcasting

Tsinkaya a nan gaba da wuya, amma tsinkaya ba da sauki.

Na biyu main dabarun amfani da masu bincike da observational data ne kiyasin. Tsinkaya a nan gaba ne notoriously wuya, amma zai iya zama mai wuce yarda da muhimmanci ga yanke shawara masu yi, ko su yi aiki a kamfanoni ko gwamnatoci.

Kleinberg et al. (2015) yayi biyu labaru cewa bayyana muhimmancin kiyasin ga wasu manufofin matsaloli. Ka yi tunanin daya siyasa mai yi, Zan kira ta Anna, wanda yana fuskantar fari da kuma dole ne yanke shawara ko don su yi ijara a shaman yi da wani irin ruwa dance ƙara da dama na ruwan sama. Wani siyasa mai yi, zan kira shi Bob, dole ne yanke shawara ko ya dauki laima aiki don kauce wa samun rigar a kan hanya gida. Dukansu Anna da Bob iya yin m yanke shawara idan sun fahimci weather, amma suna bukatar su san abubuwa daban. Anna bukatar fahimtar ko ruwan sama dance sa ruwan sama. Bob, a daya hannun, ba ya bukatar wani abu game da fahimtar causality. ya kawai bukatar wani m forecast. Social bincike sau da yawa hankali a kan abin da Kleinberg et al. (2015) kira "ruwa dance-kamar" manufofin matsaloli-mãsu mayar da hankali a kan causality-da kuma watsi da "laima-kamar" manufofin matsalolin da ake mayar da hankali a kan kiyasin.

Ina so in mayar da hankali, duk da haka, a kan wani musamman irin kiyasin kira nowcasting -A lokaci samu daga hada "yanzu" da kuma "kiyasin." Maimakon tsinkaya a nan gaba, nowcasting yunkurin hango ko hasashen ba (Choi and Varian 2012) . A wasu kalmomin, nowcasting amfani kiyasin hanyoyin matsalolin ji. Kamar yadda irin wannan, ya kamata ya kasance musamman da amfani ga gwamnatocin da suka bukata dace da m matakan game da ƙasashe. Nowcasting za a iya kwatanta mafi fili da misalin Google Mura Trends.

Ka yi tunanin cewa kana ji a bit karkashin weather haka ka rubuta "mura magunguna" a cikin wani search engine, sama a shafi na links a mayar da martani, sa'an nan kuma ya bi daya daga cikinsu zuwa m shafin yanar gizon. Yanzu kwatanta wannan aiki da ake buga daga hangen zaman gaba da search engine. Kowane lokacin, miliyoyin queries suna isa daga ko'ina cikin duniya, kuma wannan rafi na queries-abin Battelle (2006) ya kira da "database na nufi" - samar da wani kullum updated taga a cikin gama kai da duniya sani. Duk da haka, juya wannan rafi na bayani a cikin wani ji na ruwan dare na mura wuya. Kawai kirgawa up yawan queries for "mura magunguna" domin ba aiki da kyau. Ba kowa ba ne wanda ya na da mura searches for mura magunguna kuma ba kowa ke binciken for mura magunguna yana da mura.

The muhimmanci da kuma wayo zamba baya Google Mura Trends ya juya a ji matsalar a cikin wani kiyasin matsala. Amurka Cibiyar Kula da Cututtuka da kuma Rigakafin (CDC) yana da wani mura saka idanu tsarin da tattara bayanai daga likitoci a kasar. Duk da haka, daya matsalar da wannan CDC tsarin ne akwai biyu mako rahoto fada. lokacin yana daukan ga data isa daga likitoci da za a tsabtace, sarrafa, da kuma buga. Amma, a lokacin da handling wani kunno kai annoba, kiwon lafiya ofisoshin ba sa so su san nawa mura akwai makonni biyu da suka wuce. su so su san yadda mura akwai a yanzu. A gaskiya, a wasu gargajiya kafofin zamantakewa data, akwai gibba tsakanin raƙuman data tarin da kuma bayar da rahoton lags. Mai babban data kafofin, a daya hannun, ne ko da yaushe-on (Sashe 2.3.1.2).

Saboda haka, Jeremy Ginsberg da kuma abokan aiki (2009) kokarin hango ko hasashen CDC mura bayanai daga Google search data. Wannan misali ne na "tsinkaya ba" domin da masu bincike da aka ƙoƙarin gwada nawa mura akwai yanzu da tsinkaya gaba data daga CDC, nan gaba data da aka auna cikin ba. Amfani da na'ura ilmantarwa, su bincike ta miliyan 50 daban-daban search sharuddan ganin abin da ya fi gaibu na CDC mura data. Daga qarshe, suka sãmi wani sa na 45 daban-daban queries cewa jũna a zama mafi gaibu, da kuma sakamakon su quite kyau: su iya amfani da search bayanai zuwa hango ko hasashen CDC data. Bisa a sashi a kan wannan takarda, wadda aka buga a Nature, Google Mura Trends zama wani sau da yawa maimaita nasarar labarin ikon babban data.

Akwai biyu muhimmanci caveats ga wannan fili nasara, duk da haka, da kuma fahimtar wadannan caveats zai taimake ka ka kimanta, kuma suka aikata kiyasin da nowcasting. Na farko, wasan kwaikwayon na Google Mura Trends a zahiri ba yawa mafi alhẽri daga da m model cewa kiyasin yawan mura bisa mikakke extrapolation daga biyu mafi yawan 'yan ma'aunai na mura ruwan dare (Goel et al. 2010) . Kuma, a kan wani lokaci lokaci Google Mura Trends a zahiri ya fi muni da wannan mai sauki m (Lazer et al. 2014) . A wasu kalmomin, Google Mura Trends da dukan bayanai, inji ilmantarwa, da kuma iko kwamfuta bai cika fuska outperform mai sauki da kuma sauki su fahimci heuristic. Wannan ya nuna cewa a lokacin da kimantawa wani forecast ko nowcast yana da muhimmanci kwatanta da a baseline.

Na biyu da muhimmanci gargadi game da Google Mura Trends shi ne cewa da ikon hango ko hasashen CDC mura data kasance yiwuwa ga gajere gazawar da kuma dogon lokaci lalace saboda gantali da algorithmic confounding. Alal misali, a lokacin da 2009 alade mura fashewa Google Mura Trends da cika fuska a kan-kiyasta adadin mura, watakila saboda mutane sukan canza su search hali a mayar da martani ga tartsatsi tsoron a duniya cutar AIDS (Cook et al. 2011; Olson et al. 2013) . Bugu da kari wadannan gajere matsaloli, wasan kwaikwayon hankali rududdugaggu a kan lokaci. Diagnosing cikin dalilan wannan dogon lokaci lalace ne m saboda Google search Algorithms ne mallakar tajirai, amma ya bayyana cewa, a shekarar 2011 Google yi canje-canje da zai bayar da shawarar da alaka search sharuddan a lokacin da mutane bincika cututtuka kamar "zazzabi" da kuma "tari" (shi ma ze cewa wannan alama ne ba aiki). Ƙara wannan alama ne a kaucewa m abu ya yi idan kana gudu a search engine business, kuma shi yana da sakamako na samar da mafi kiwon lafiya ya shafi bincike. Wannan shi yiwuwa wani rabo ga harkokin kasuwanci, amma shi ya sa Google Mura Trends zuwa kan-kimanta mura ruwan dare (Lazer et al. 2014) .

Abin farin, wadannan matsaloli tare da Google Mura Trends ne fixable. A gaskiya ma, ta yin amfani da more m hanyoyin, Lazer et al. (2014) da kuma Yang, Santillana, and Kou (2015) sun iya samun mafi alhẽri results. Going ido, ina sa ran cewa nowcasting karatu da hada babban data da bincike tattara data-da hada Duchamp-style Readymades da Michaelangelo-style Custommades-zai taimaka manufofin masu samar da sauri kuma mafi m ma'aunai na yanzu da kuma tsinkaya na nan gaba.