{"id":1024,"date":"2018-02-12T19:32:17","date_gmt":"2018-02-12T16:32:17","guid":{"rendered":"http:\/\/www.gokberkcan.com\/?p=1024"},"modified":"2020-11-18T20:42:12","modified_gmt":"2020-11-18T17:42:12","slug":"rda-panel-veri-analizi","status":"publish","type":"post","link":"https:\/\/www.gokberkcan.com\/?p=1024","title":{"rendered":"R'da panel veri analizi"},"content":{"rendered":"\n

R’da panel veri analizini Google’da arad\u0131\u011f\u0131n\u0131zda \u00e7\u0131kan T\u00fcrk\u00e7e sonu\u00e7lar \u00e7o\u011funlukla panel veriyi anlat\u0131yor. \u0130ngilizce kaynaklarda (kitap, dergi, forum, paketler) son derece fazla payla\u015f\u0131lan R’da panel veri analizi i\u00e7in bir de T\u00fcrk\u00e7e kaynak gerekti\u011fi fikrinden yola \u00e7\u0131k\u0131p bu yaz\u0131y\u0131 haz\u0131rlad\u0131m. Ekonometrik \u00e7al\u0131\u015fmalar\u0131n\u0131 R’da yapmak isteyenlere en ufak bir katk\u0131m olduysa ne mutlu bana.<\/span><\/p>\n\n\n\n

\u0130htiyac\u0131m\u0131z olanlar:<\/span><\/p>\n\n\n\n

  1. R<\/a> (bizzat kendisi)<\/span><\/li>
  2. RStudio<\/a> (veri aktar\u0131m\u0131n\u0131 kolayla\u015ft\u0131rmak i\u00e7in)<\/span><\/li>
  3. plm, lmtest ve sandwich paketleri<\/span><\/li><\/ol>\n\n\n\n

    Devam\u0131 a\u015fa\u011f\u0131da<\/span><\/p>\n\n\n\n

    <\/span><\/wp-block><\/p>\n\n\n\n

    Yap\u0131lacaklar<\/span><\/p>\n\n\n\n

    R ve RStudio kurulumlar\u0131 bittikten sonra RStudio’da a\u015fa\u011f\u0131daki komutlar\u0131 \u00e7al\u0131\u015ft\u0131r\u0131n.<\/span><\/p>\n\n\n\n

    install.packages(“plm”)<\/span>
    install.packages(“sandwich”)<\/span>
    install.packages(“lmtest”)<\/span>
    library(plm)<\/span>
    library(sandwich)<\/span>
    library(lmtest)<\/span><\/p><\/blockquote>\n\n\n\n

    Hangi paketi neden kuruyoruz?<\/span><\/p>\n\n\n\n

    Panel veri setinizi analiz etmek i\u00e7in “plm” kurman\u0131z gerekli.<\/span><\/p>\n\n\n\n

    Ekonometrik testlerin sonu\u00e7lar\u0131na g\u00f6re de tahmin vas\u0131tas\u0131 (estimator) \u00e7e\u015fitlerini artt\u0131rmak i\u00e7in de “sandwich” paketine ihtiyac\u0131m\u0131z var.<\/span><\/p>\n\n\n\n

    Varyans oynakl\u0131\u011f\u0131 testini kolayla\u015ft\u0131rmak i\u00e7in de “lmtest” paketini kullanaca\u011f\u0131z.<\/span><\/p>\n\n\n\n

    Hangi veri \u00fcst\u00fcnde analiz yapaca\u011f\u0131z?<\/span><\/p>\n\n\n\n

    E\u011fer elinizde panel olarak haz\u0131rlanm\u0131\u015f bir veriseti varsa onun \u00fcst\u00fcnde \u00e7al\u0131\u015fabilirsiniz. Sadece idman yapmak i\u00e7in R’da mevcut bir seti \u00fcst\u00fcnden de analizkeri yapabilirsiniz. Anlat\u0131m\u0131 kolayla\u015ft\u0131rmak i\u00e7in ben plm ile gelen “Gasoline” verisini (4 de\u011fi\u015fken, 18 birim, 19 y\u0131l, 342 g\u00f6zlem) vekt\u00f6r i\u00e7inde kullanaca\u011f\u0131m.<\/span><\/p>\n\n\n\n

    data(“Gasoline”)<\/span>
    dene<-plm(lgaspcar~lincomep+lrpmg+lcarpcap,data=Gasoline)<\/span><\/p><\/blockquote>\n\n\n\n

    En temel haliyle panel veri setinin analizidir yukar\u0131daki  “dene” vekt\u00f6r\u00fc ve \u00e7al\u0131\u015ft\u0131rd\u0131\u011f\u0131n\u0131zda tahmin \u00e7arpanlar\u0131n\u0131 g\u00f6sterir R. plm() aynen lm() fonksiyonu gibi \u00e7al\u0131\u015f\u0131r yani siz ba\u011f\u0131ml\u0131 de\u011fi\u015fken (y) ile ba\u011f\u0131ms\u0131z de\u011fi\u015fkenleri (x) tan\u0131mlaman\u0131z (y~x1+x2+x3+…+xn) gereklidir. E\u011fer modele ili\u015fkin test sonu\u00e7lar\u0131n\u0131 g\u00f6rmek istiyorsan\u0131z vekt\u00f6r\u00fc summary() fonksiyonu ile \u00e7al\u0131\u015ft\u0131rman\u0131z gerekli.<\/span><\/p>\n\n\n\n

    summary(dene) <\/span><\/p><\/blockquote>\n\n\n\n

    Peki panel regresyonunuzu etkilerin (havuzland\u0131r\u0131lm\u0131\u015f, rassal veya sabit), etki t\u00fcrlerinin (zaman, birim, ikiy\u00f6nl\u00fc, yerle\u015fik), endekslerin ve gerekirse tahmin vas\u0131talar\u0131n\u0131n tan\u0131mlad\u0131\u011f\u0131 \u015fekilde geni\u015fletmek istersek vekt\u00f6r\u00fc a\u015fa\u011f\u0131daki haliyle yazmak gerekir.<\/span><\/p>\n\n\n\n

    dene<-plm(lgaspcar~lincomep+lrpmg+lcarpcap,data=Gasoline, index = c(“country”, “year”), model = “”, effect=””, random.method = “”<\/strong>)<\/span><\/p><\/blockquote>\n\n\n\n

    model= <\/strong>kriteri pooling (klasik OLS), within (sabit), random (rassal), between, fd (ilk farklar), ht (Hausman-Taylor tahmin vas\u0131tas\u0131) se\u00e7eneklerinden biriyle<\/span><\/p>\n\n\n\n

    • pooling: havuzland\u0131r\u0131lm\u0131\u015f etkiler, klasik OLS<\/span><\/li>
    • within: sabit etkiler<\/span><\/li>
    • random: rassal etkiler<\/span><\/li>
    • fd: ilk farklar<\/span><\/li>
    • ht: Hausman-Taylor tahmin vas\u0131tas\u0131<\/span><\/li>
    • between: orta tahmincisi<\/span><\/li><\/ul>\n\n\n\n

      effect= <\/strong>kriteri individual, time, twoways, nested se\u00e7eneklerinden biriyle<\/span><\/p>\n\n\n\n

      • individual: birim etkisi<\/span><\/li>
      • time: zaman etkisi<\/span><\/li>
      • twoways: iki y\u00f6nl\u00fc etki<\/span><\/li>
      • nested: yerle\u015fik etki<\/span><\/li><\/ul>\n\n\n\n

        random.method=<\/strong> kriteri sadece model etkisi olarak “random” se\u00e7ildi\u011finde kullan\u0131labilir ve a\u015fa\u011f\u0131daki tahmincilerinin se\u00e7ilmesini sa\u011flar.<\/span><\/p>\n\n\n\n

        • swar (tahminci se\u00e7ilmezse plm taraf\u0131ndan varsay\u0131lan olarak kullan\u0131l\u0131r)<\/span><\/li>
        • amemiya<\/span><\/li>
        • walhus<\/span><\/li>
        • nerlove<\/span><\/li><\/ul>\n\n\n\n

          Panel regresyonunuzdaki model etkileri i\u00e7in gerekli testleri yine plm paketinin alt\u0131ndaki komutlarla \u00f6l\u00e7ebilirsiniz.<\/span><\/p>\n\n\n\n

          Panel veri testlerini nas\u0131l y\u00fcr\u00fctece\u011fiz?<\/span><\/p>\n\n\n\n

          Havuzland\u0131r\u0131lm\u0131\u015f ile Rassal etkiler aras\u0131nda karar vermek i\u00e7in Lagrange Multiplier testini yapaca\u011f\u0131z.Rassal ile Sabit etkiler aras\u0131nda karar verebilmek i\u00e7in de Hausman (1978) testi yapaca\u011f\u0131z. Bunun i\u00e7in modelinizi s\u0131ras\u0131yla (pooled, random, within) <\/strong>ile kaydedin.<\/span><\/p>\n\n\n\n

          dene.pe<-plm(lgaspcar~lincomep+lrpmg+lcarpcap,data=Gasoline, index = c(“country”, “year”), model = “pooling”<\/strong>)<\/span>
          dene.re<-plm(lgaspcar~lincomep+lrpmg+lcarpcap,data=Gasoline, index = c(“country”, “year”), model = “random”<\/strong>)<\/span>
          dene.fe<-plm(lgaspcar~lincomep+lrpmg+lcarpcap,data=Gasoline, index = c(“country”, “year”), model = “within”<\/strong>)<\/span><\/p><\/blockquote>\n\n\n\n

          Sonras\u0131nda ise her bir modeli teker teker plmtest() fonksiyonu ile deneyin ve p-de\u011ferlerine g\u00f6re kullanabilece\u011finiz etki t\u00fcr\u00fcn\u00fc g\u00f6rm\u00fc\u015f olacaks\u0131n\u0131z.<\/span><\/p>\n\n\n\n

          plmtest(dene.pe)<\/span>
          plmtest(dene.re)<\/span>
          plmtest(dene.fe)<\/span><\/p><\/blockquote>\n\n\n\n

          Peki modelin i\u00e7erisindeki birim, zaman veya iki y\u00f6nl\u00fc etkileri nas\u0131l \u00f6l\u00e7ebiliriz? Onu da yine plmtest() fonksiyonu yard\u0131m\u0131yla ger\u00e7ekle\u015ftirece\u011fiz. <\/span><\/p>\n\n\n\n

          plmtest(dene, effect =<\/strong> “”, type=<\/strong>“”)<\/span><\/p><\/blockquote>\n\n\n\n

          effect= <\/strong>kriteri individual, time, twoways, se\u00e7eneklerinden biriyle<\/span><\/p>\n\n\n\n

          • individual: birim etkisi<\/span><\/li>
          • time: zaman etkisi<\/span><\/li>
          • twoways: iki y\u00f6nl\u00fc etki<\/span><\/li><\/ul>\n\n\n\n

            type=<\/strong> kriteri honda, bp, kw, ghm test se\u00e7eneklerinden biriyle <\/span><\/p>\n\n\n\n

            • honda: Honda (1985) (se\u00e7enek belirtilmezse varsay\u0131lan olarak kullan\u0131l\u0131r.)<\/span><\/li>
            • bp: Breusch\/Pagan (1980)<\/span><\/li>
            • kw: King\/Wu (1997)<\/span><\/li>
            • ghm: Gourieroux\/Holly\/Monfort (1982);<\/span><\/li><\/ul>\n\n\n\n

              Birka\u00e7 farkl\u0131 fonksiyon ile de modelin i\u00e7indeki sorunlar\u0131 test edece\u011fiz. Test sonu\u00e7lar\u0131n\u0131n p-de\u011ferine g\u00f6re de sorunlar\u0131n varl\u0131\u011f\u0131na g\u00f6re tahmin vas\u0131talar\u0131 kullanaca\u011f\u0131z.<\/span><\/p>\n\n\n\n

              Breusch-Godfrey\/Wooldridge seriler aras\u0131 korelasyon testi ile ba\u015flayal\u0131m. (Breusch, 1978; Godfrey, 1978; Wooldridge, 2013)\u2060 <\/span><\/p>\n\n\n\n

              pbgtest(dene)<\/span><\/p><\/blockquote>\n\n\n\n

              p-testi < 0.05 => Seriler aras\u0131 korelasyon vard\u0131r.<\/em><\/span><\/p>\n\n\n\n

              p-testi > 0.05 => Seriler aras\u0131 korelasyon yoktur.<\/em><\/span><\/p>\n\n\n\n

              Yatay kesit ba\u011f\u0131ml\u0131\u011f\u0131 Pesaran (2004) ve Breusch-Pagan LM korelasyon testlerini uygulayaca\u011f\u0131z.<\/span><\/p>\n\n\n\n

              pcdtest(dene, test=”cd”)<\/span>
              pcdtest(dene, test=”lm”)<\/span><\/p><\/blockquote>\n\n\n\n

              p-testi < 0.05 => Yatay kesit ba\u011f\u0131ml\u0131\u011f\u0131 vard\u0131r.<\/em><\/span><\/p>\n\n\n\n

              p-testi > 0.05 =>Yatay kesit ba\u011f\u0131ml\u0131\u011f\u0131 yoktur.<\/em><\/span><\/p>\n\n\n\n

              Varyans oynakl\u0131\u011f\u0131 (Heteroskedasticity) sorununun varl\u0131\u011f\u0131n\u0131n test edilmesi i\u00e7in “lmtest” paketinden destek alaca\u011f\u0131z. Bunun i\u00e7in \u00f6nce modeli lm() fonksiyonu ile vekt\u00f6r olarak kaydedece\u011fiz.<\/span><\/p>\n\n\n\n

              dene.lm<-lm(lgaspcar~lincomep+lrpmg+lcarpcap,data=Gasoline)<\/span><\/p><\/blockquote>\n\n\n\n

              dene.lm vekt\u00f6r\u00fcn\u00fc de lmtest paketinin i\u00e7erisinde mevcut bulunan bptest() fonksiyonu ile test edece\u011fiz.<\/span><\/p>\n\n\n\n

              bptest(dene.lm, studentize = F)<\/span><\/p><\/blockquote>\n\n\n\n

              p-testi < 0.05 => Varyans oynakl\u0131\u011f\u0131 vard\u0131r.<\/em><\/span><\/p>\n\n\n\n

              p-testi > 0.05 =>Varyans oynakl\u0131\u011f\u0131 yoktur.<\/em><\/span><\/p>\n\n\n\n

              S\u0131ra geldi “sandwich”ten at\u0131\u015ft\u0131rmaya<\/span><\/p>\n\n\n\n

              Yapt\u0131\u011f\u0131n\u0131z testlerin \u00e7e\u015fitlili\u011fine g\u00f6re kullanman\u0131z gereken tahmin vas\u0131talar\u0131 mevcut, plm paketinin de destek ald\u0131\u011f\u0131 “sandwich” \u00fczerinden \u00e7al\u0131\u015faca\u011f\u0131z.<\/span><\/p>\n\n\n\n

              Modelinizde \u00e7\u0131kan sorunlar\u0131n \u00e7e\u015fitlili\u011fine g\u00f6re varyans d\u00fczeltmeleri yapman\u0131z gerekiyor. Bunun i\u00e7in de vcov ile ba\u015flayan fonksiyonlar \u00fczerinden \u00e7al\u0131\u015faca\u011f\u0131z.<\/span><\/p>\n\n\n\n

              • vcovSCC() Yatay kesit ve seriler aras\u0131 korelasyon oldu\u011fu durumlarda Driscoll&Kraay tahmin vas\u0131tas\u0131 i\u00e7in<\/span><\/li>
              • vcovHC(): Varyans oynakl\u0131\u011f\u0131n\u0131 \u00e7\u00f6zmede White tahmin vas\u0131tas\u0131 i\u00e7in<\/span><\/li>
              • vcovHAC(): Varyans oynakl\u0131\u011f\u0131 ve otokorelasyon uyumlu (Heteroskedasticity and autocorrelation consistent [HAC]) tahmin vas\u0131tas\u0131 i\u00e7in <\/span><\/li>
              • vcovNW(): Seriler aras\u0131 korelasyon sorununu \u00e7\u00f6zmede Newey-West tahmin vas\u0131tas\u0131 kullanmak i\u00e7in <\/span><\/li>
              • vcovPL(): Newey-West ve Driscoll&Kraay tahmin vas\u0131talar\u0131n\u0131 ayn\u0131 anda kullanmak i\u00e7in <\/span><\/li><\/ul>\n\n\n\n

                vcov fonksiyonu ile varyans testi yapabilece\u011finiz gibi plm taraf\u0131ndan desteklenen fonksiyonlar\u0131 summary() i\u00e7inde kullanarak modelin tahmin vas\u0131tas\u0131 ile sa\u011flamla\u015ft\u0131r\u0131lm\u0131\u015f testlerini de yapabilirsiniz.<\/span><\/p>\n\n\n\n

                summary(dene.fe, vcov = vcovSCC)<\/span>
                summary(dene.re, vcov = vcovNW)<\/span>
                summary(dene.pe, vcov = vcovHC)<\/span><\/p><\/blockquote>\n\n\n\n

                Ekonometri bilginizin derinli\u011fine g\u00f6re plm ve di\u011fer paketleri daha detayl\u0131 kullanman\u0131z hatta geli\u015ftirmeniz dahi m\u00fcmk\u00fcn. Derdini anlatacak kadar ekonometri bilen bendenize  de bir \u015feyler anlat\u0131rsan\u0131z mutlu olurum.<\/span><\/p>\n\n\n\n

                Kaynak\u00e7a<\/strong><\/span><\/p>\n\n\n\n

                Achim Zeileis, Torsten Hothorn (2002). Diagnostic Checking in Regression<\/span>
                Relationships. R News 2(3), 7-10. URL https:\/\/CRAN.R-project.org\/doc\/Rnews\/<\/span><\/p>\n\n\n\n

                Achim Zeileis (2004). Econometric Computing with HC and HAC Covariance Matrix<\/span>
                Estimators. Journal of Statistical Software 11(10), 1-17. URL http:\/\/www.jstatsoft.org\/v11\/i10\/.<\/span><\/p>\n\n\n\n

                Achim Zeileis (2006). Object-Oriented Computation of Sandwich Estimators. Journal of Statistical Software 16(9), 1-16. URL http:\/\/www.jstatsoft.org\/v16\/i09\/.<\/span><\/p>\n\n\n\n

                Croissant Y and Millo G (2008). \u201cPanel Data Econometrics in R: The plm Package.\u201d<\/span>
                _Journal of Statistical Software_, *27*(2), pp. 1-43. doi: 10.18637\/jss.v027.i02 (URL:<\/span>
                http:\/\/doi.org\/10.18637\/jss.v027.i02).<\/span><\/p>\n\n\n\n

                Millo G (2017). \u201cRobust Standard Error Estimators for Panel Models: A Unifying<\/span>
                Approach.\u201d _Journal of Statistical Software_, *82*(3), pp. 1-27. doi:<\/span>
                10.18637\/jss.v082.i03 (URL: http:\/\/doi.org\/10.18637\/jss.v082.i03).<\/span><\/p>\n\n\n\n

                R Core Team. (2017). R: A language and environment for statistical computing. R<\/span>
                Foundation for Statistical Computing, Vienna, Austria. URL<\/span>
                https:\/\/www.R-project.org\/.<\/span><\/p>\n\n\n\n

                RStudio Team. (2016). RStudio: Integrated Development for R. RStudio, Inc., Boston, MA<\/span>
                URL http:\/\/www.rstudio.com\/.<\/span><\/p>\n\n\n\n

                Torres-Reyna, Oscar. (2010) https:\/\/www.princeton.edu\/~otorres\/Panel101R.pdf<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

                R’da panel veri analizini Google’da arad\u0131\u011f\u0131n\u0131zda \u00e7\u0131kan T\u00fcrk\u00e7e sonu\u00e7lar \u00e7o\u011funlukla panel veriyi anlat\u0131yor. \u0130ngilizce kaynaklarda (kitap, dergi, forum, paketler) son derece fazla payla\u015f\u0131lan R’da panel veri analizi i\u00e7in bir de T\u00fcrk\u00e7e kaynak gerekti\u011fi fikrinden yola \u00e7\u0131k\u0131p bu yaz\u0131y\u0131 haz\u0131rlad\u0131m. Ekonometrik \u00e7al\u0131\u015fmalar\u0131n\u0131 R’da yapmak isteyenlere en ufak bir katk\u0131m olduysa ne mutlu bana. \u0130htiyac\u0131m\u0131z olanlar: R […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[9],"tags":[],"class_list":["post-1024","post","type-post","status-publish","format-standard","hentry","category-r"],"_links":{"self":[{"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=\/wp\/v2\/posts\/1024","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1024"}],"version-history":[{"count":1,"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=\/wp\/v2\/posts\/1024\/revisions"}],"predecessor-version":[{"id":1938,"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=\/wp\/v2\/posts\/1024\/revisions\/1938"}],"wp:attachment":[{"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1024"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1024"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.gokberkcan.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1024"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}