{"id":24079,"date":"2025-10-15T18:57:10","date_gmt":"2025-10-15T16:57:10","guid":{"rendered":"https:\/\/varify.io\/?p=24079"},"modified":"2025-10-19T18:18:31","modified_gmt":"2025-10-19T16:18:31","slug":"multivariate-tests","status":"publish","type":"post","link":"https:\/\/varify.io\/tr\/blog\/cok-degiskenli-testler\/","title":{"rendered":"\u00c7ok de\u011fi\u015fkenli testler: Neden (genellikle) k\u00f6t\u00fc bir fikirdirler - ve ne zaman ger\u00e7ekten faydal\u0131 olurlar"},"content":{"rendered":"
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

\u00c7ok de\u011fi\u015fkenli testler (MVT) ilk bak\u0131\u015fta g\u00fc\u00e7l\u00fc g\u00f6r\u00fcn\u00fcr - birka\u00e7 \u00f6\u011feyi ayn\u0131 anda test edebilir ve etkile\u015fimlerini tan\u0131yabilirsiniz.<\/p>

Ancak pratikte, trafi\u011fe a\u00e7<\/b>, istatistiksel olarak karma\u015f\u0131k<\/b> ve \u00e7o\u011fu web sitesi i\u00e7in \u00e7ok yava\u015f<\/b>.<\/p>

MVT yaln\u0131zca ayda birka\u00e7 y\u00fcz bin ziyaret\u00e7i<\/b> veya 5%<\/b> \u00fczerindeki d\u00f6n\u00fc\u015f\u00fcm oranlar\u0131 i\u00e7in anlaml\u0131d\u0131r.<\/p>

O zamana kadar: \u0130yi planlanm\u0131\u015f A\/B\/n testleri ge\u00e7erli bulgular\u0131 daha h\u0131zl\u0131 sunar<\/strong>.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

\u00c7ok de\u011fi\u015fkenli testler asl\u0131nda ne yapar?<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

\u00c7ok de\u011fi\u015fkenli bir test \u00e7e\u015fitli unsurlar\u0131 ayn\u0131 anda<\/strong> test eder.<\/p>

\u00d6rnek:
2 ba\u015fl\u0131k \u00d7 2 resim \u00d7 2 d\u00fc\u011fme test edersiniz \u2192 bu 8 kombinasyon<\/strong> ile sonu\u00e7lan\u0131r.<\/p>

Her kombinasyon trafikten pay\u0131n\u0131 al\u0131r - e\u015fit olarak da\u011f\u0131t\u0131l\u0131rsa ziyaret\u00e7ilerin yaln\u0131zca 1\/8'i.<\/p>

\u0130\u015fte sorun da tam olarak burada yat\u0131yor: Varyant ba\u015f\u0131na trafik dramatik bir \u015fekilde azal\u0131yor<\/strong> ve bununla birlikte verilerinizin anlaml\u0131l\u0131\u011f\u0131<\/strong> da azal\u0131yor.<\/p>

Sonu\u00e7: Testler \u00e7ok daha uzun<\/strong> s\u00fcrer, anlaml\u0131l\u0131\u011fa daha ge\u00e7<\/strong> ula\u015f\u0131l\u0131r ve bu arada di\u011fer deneyleri<\/strong> engellersiniz.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

\u00c7ok de\u011fi\u015fkenli testler neden nadiren faydal\u0131d\u0131r?<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
\n\t\t\t<\/i>\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

1. Trafik \u00e7ok fazla de\u011fi\u015fkene yay\u0131lm\u0131\u015f durumda<\/h3>\nDaha fazla kombinasyon, ziyaret\u00e7ilerinizin bir\u00e7ok gruba yay\u0131lmas\u0131 anlam\u0131na gelir.\n Bu da istatistiksel g\u00fcc\u00fc (yani ger\u00e7ek etkileri fark etme kabiliyetini) azalt\u0131r.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
\n\t\t\t<\/i>\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

2. testler sonsuza kadar s\u00fcrer<\/h3>

Ayl\u0131k 100.000 ziyaret\u00e7iyle bile bir sonu\u00e7 g\u00f6rmeniz aylar<\/strong> alabilir - oysa basit bir A\/B testiyle uzun zaman \u00f6nce birka\u00e7 ge\u00e7erli bilgi elde edebilirdiniz.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
\n\t\t\t<\/i>\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

3. daha karma\u015f\u0131k istatistikler ve daha y\u00fcksek hata oran\u0131<\/h3>

Her ek varyant rastgele yanl\u0131\u015f bir kazanan bulma<\/strong> riskinizi art\u0131r\u0131r (buna alfa hata birikimi<\/strong> denir).<\/p>

Ne kadar \u00e7ok test yaparsan\u0131z, bir varyant\u0131n \u00f6yle olmad\u0131\u011f\u0131 halde yanl\u0131\u015fl\u0131kla daha iyi<\/strong> g\u00f6r\u00fcnme ihtimali o kadar artar. Bu ayn\u0131 zamanda \"\u00e7oklu kar\u015f\u0131la\u015ft\u0131rma sorunu\" olarak da bilinir.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
\n\t\t\t<\/i>\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

4 K\u00fc\u00e7\u00fck etkiler, b\u00fcy\u00fck \u00e7aba<\/h3>\nMVT'ler genellikle yaln\u0131zca ince ayar etkilerini g\u00f6sterir - \u00f6rne\u011fin A d\u00fc\u011fmesinin B ba\u015fl\u0131\u011f\u0131yla minimum d\u00fczeyde daha iyi performans g\u00f6stermesi gibi. Uygulamada, bu t\u00fcr mini etkiler genellikle \u00e7abay\u0131 hakl\u0131 \u00e7\u0131karacak kadar \u00f6nemli de\u011fildir.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

\"G\u00fc\u00e7\" ger\u00e7ekten ne anlama geliyor (basit\u00e7e a\u00e7\u0131klanm\u0131\u015ft\u0131r)<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Bir testin g\u00fcc\u00fc<\/strong>, ger\u00e7ekte ger\u00e7ek bir etkiyi<\/strong> tespit etme olas\u0131l\u0131\u011f\u0131n\u0131z\u0131 tan\u0131mlar.<\/p>

\u00d6rne\u011fin, bir testin g\u00fcc\u00fc 80 %<\/strong> ise, bu, ger\u00e7ek bir fark varsa, bunu istatistiksel olarak da 10 vakan\u0131n 8'inde<\/strong> bulaca\u011f\u0131n\u0131z anlam\u0131na gelir.<\/p>

Her 10 vakadan 2'sinde (20 %) yanl\u0131\u015fl\u0131kla hi\u00e7bir etki olmad\u0131\u011f\u0131na<\/strong> inan\u0131rs\u0131n\u0131z.<\/p>

Y\u00fckseltme ne kadar k\u00fc\u00e7\u00fck olursa, ayn\u0131 g\u00fc\u00e7 de\u011ferini korumak i\u00e7in daha fazla ziyaret\u00e7iye<\/strong> ihtiyac\u0131n\u0131z olur.<\/p>

Bu nedenle 8 kombinasyonlu bir MVT, basit bir A\/B testi ile ayn\u0131 g\u00fcce ula\u015fmak i\u00e7in \u00e7ok daha fazla trafi\u011fe<\/strong> ihtiya\u00e7 duyar.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Neden varyant ba\u015f\u0131na en az 500 d\u00f6n\u00fc\u015f\u00fcme ihtiyac\u0131n\u0131z var?<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

CRO uygulamas\u0131n\u0131n kaba bir kural\u0131 \u015fudur:<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t

\n\t\t\t\tMakul \u00f6l\u00e7\u00fcde istikrarl\u0131 bir a\u00e7\u0131klama yapabilmek i\u00e7in varyant ba\u015f\u0131na en az 500 d\u00f6n\u00fc\u015f\u00fcm gereklidir.\t\t\t<\/p>\n\t\t\t\t\t<\/blockquote>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Bunun \u00fc\u00e7 nedeni vard\u0131r:<\/p>

  1. D\u00f6n\u00fc\u015f\u00fcm oranlar\u0131 \u00e7ok k\u00fc\u00e7\u00fck \u00f6rneklerde a\u015f\u0131r\u0131 derecede dalgalan\u0131r - k\u00fc\u00e7\u00fck tesad\u00fcfler sonucu bozabilir.<\/li>
  2. Varyant ba\u015f\u0131na olay say\u0131s\u0131 \u00e7ok k\u00fc\u00e7\u00fckse ara\u00e7lar temiz anlaml\u0131l\u0131k testleri yapamaz.<\/li>
  3. Yakla\u015f\u0131k 500 d\u00f6n\u00fc\u015f\u00fcmden itibaren da\u011f\u0131l\u0131m sabitle\u015fir, b\u00f6ylece k\u00fc\u00e7\u00fck farklar bile g\u00f6r\u00fcn\u00fcr hale gelir.<\/li><\/ol>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\t

    \u00d6nemli:<\/strong> Bu, varyant ba\u015f\u0131na<\/strong> - yani bir MVT'deki her kombinasyon i\u00e7in de ge\u00e7erlidir.<\/p>

    8 kombinasyonunuz varsa, sadece temel gereksinimi kar\u015f\u0131lamak i\u00e7in en az 8 \u00d7 500 = 4000 d\u00f6n\u00fc\u015f\u00fcme<\/strong> ihtiyac\u0131n\u0131z vard\u0131r.<\/p>

    D\u00f6n\u00fc\u015f\u00fcm oran\u0131na ba\u011fl\u0131 olarak, bu h\u0131zl\u0131 bir \u015fekilde y\u00fcz binlerce ziyaret\u00e7i<\/strong> \u00e7ekebilir.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

    \n\t\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t

    Alfa hatas\u0131: Neden daha fazla varyant riski art\u0131r\u0131r?<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\t

    alpha hatas\u0131 (birinci t\u00fcr hata)<\/strong>, ger\u00e7ek bir fark olmamas\u0131na ra\u011fmen yanl\u0131\u015fl\u0131kla bir varyant\u0131 kazanan olarak tan\u0131mlama<\/strong> olas\u0131l\u0131\u011f\u0131n\u0131 tan\u0131mlar.<\/p>

    Bir \u00f6rnek:
    5 % (\u03b1=0.05\\alpha=0{,}05\u03b1=0.05) anlaml\u0131l\u0131k d\u00fczeyi ile \u00e7al\u0131\u015f\u0131yorsan\u0131z, bu \u015fu anlama gelir<\/p>