ã¯ããã«
å¾åã¹ã³ã¢ã«é¢ãã¦æãã¨ãããä»åã¯è¨äºã«ãã¦ããããã¨æãã¾ãã
å¶ã«ãå¾åã¹ã³ã¢ã«é¢ãã¦twitterã§åããã¨ãããã®ã§ãããå¤ãã®äººãæ±ãã¦ãå¾åã¹ã³ã¢ãæã¤ã¹ãæ§è³ªã«é¢ãã¦åéããããã®ã§ã¯ãªããã¨æãããã®è¨äºãæ¸ãã«è³ãã¾ããã
å¾åã¹ã³ã¢ã®ç®ç
å¾åã¹ã³ã¢ã¯ã©ã®ãããªç®çã§å°åºããã¦ããããçæ§ã¯ç解ãã¦ããã§ããããï¼ç§ãå¤ãè¦èãããã®ã¯ããå¾åã¹ã³ã¢ã¨ã¯ä»å ¥ç¢ºçã§ãããä»å ¥ãè¡ããã確çã精度ããäºæ¸¬ã§ããã¢ãã«ãæ§ç¯ãããã®ç¢ºçãå¾åã¹ã³ã¢ã¨å¼ã¶ã¹ãã ãã¨ãã£ãè¨èª¬ã§ããå½¢å¼ä¸ãä»å ¥ç¾¤ã1ãå¯¾ç §ç¾¤ã0ã¨ãã1ã¨ãªã確çãäºæ¸¬ããã¢ãã«ãçµã¿ããã®ã¢ãã«ã«ãã£ã¦äºæ¸¬ããã確çå¤ãå¾åã¹ã³ã¢ã¨å¼ã¶ããã§ããããåç´ã«èããã°ããã®ãããªè¦è§£ãåºã¾ãã®ã¯ç解ã§ãã¾ãã
ããããå¾åã¹ã³ã¢ã®æ¬æ¥ã®ç®çã¯ããã«ã¯ããã¾ãããæ¬è¨äºã§ã¯ãå¾åã¹ã³ã¢ãããã³ã°ã«ç¦ç¹ãå½ã¦ã¤ã¤ãå¾åã¹ã³ã¢ãæ¬æ¥æã¤ã¹ãç¹æ§ã«ã¤ãã¦è¿°ã¹ã¦ããããã¨æãã¾ãã
ç§ã¯ãå¾åã¹ã³ã¢ã®æ¬æ¥ã®ç®çã¯å ±å¤éåå¸ã®è¦ç´å¤ã§ããã¨èãã¦ãã¾ããããã¦ãå¾åã¹ã³ã¢ãããã³ã°ã®ç®çã¯ãå¾åã¹ã³ã¢ã«ãã£ã¦é常ã«å ±å¤éåå¸ã®è¿ããå¯¾ç §ç¾¤ãè¦ã¤ããä»å ¥ç¾¤1人ã«å¯¾ãã¦ã®æ½å¨çµæï¼ä»å ¥ãè¡ãããªãã£ãå ´åã®çµæï¼ãé©åã«æ¨å®ãããã¨ã§ããã¨è§£éãã¦ãã¾ãããã®è§£éã¯ãIPWï¼Inverse Probability Weightingï¼ãªã©ä»ã®ææ³ã«ããã¦ãå¾åã¹ã³ã¢ãå ±å¤éåå¸ã®è¦ç´å¤ã§ãããã¨ãåæã¨ãã¦ããç¹ã¨ä¸è´ãã¾ãã
å¾åã¹ã³ã¢ã®æ¬è³ªï¼å ±å¤éåå¸ã®è¦ç´å¤
Rosenbaumã¨Rubinï¼1983ï¼ã¯æ¬¡ã®ããã«è¿°ã¹ã¦ãã¾ãï¼
"The propensity score is a balancing score: given the propensity score, the distribution of observed baseline covariates will be similar between treated and control units." ï¼å¾åã¹ã³ã¢ã¯ãã©ã³ã·ã³ã°ã¹ã³ã¢ã§ãããå¾åã¹ã³ã¢ãä¸ããããã¨ãä»å ¥ç¾¤ã¨å¯¾ç §ç¾¤ã®è¦³æ¸¬ãããåºæºå ±å¤éã®åå¸ã¯é¡ä¼¼ãããï¼
ãã®ããã©ã³ã·ã³ã°ããããã£ï¼Balancing Propertyï¼ããå¾åã¹ã³ã¢ã®æ¬è³ªã§ããå ·ä½çã«ã¯ã以ä¸ã®æ¡ä»¶ãæãç«ã¡ã¾ãï¼
ããªãã¡ãå¾åã¹ã³ã¢ ã«æ¡ä»¶ä»ãããå ´åãä»å ¥ã®æç¡ ã¨å ±å¤é ã¯ç¬ç«ã¨ãªãã¾ãããã®ç¹æ§ã«ãããå¾åã¹ã³ã¢ãå©ç¨ããææ³ããã¤ã¢ã¹ãåæ¸ããå æå¹æã®æ¨å®ãå¯è½ã«ãã¾ãã
ã¾ãããã®ç¹æ§ãéè¦ã§ããçç±ã¯ãç¾å®ã®è¦³æ¸¬ãã¼ã¿ã§ã¯å ±å¤éåå¸ãä»å ¥ç¾¤ã¨å¯¾ç §ç¾¤ã§ãã°ãã°ç°ãªãããã§ãããã®å·®ç°ãããéããä»å ¥å¹æã®æ¨å®ã«ã¯ãã¤ã¢ã¹ãçãã¾ããå¾åã¹ã³ã¢ãç¨ãããã¨ã§ãå ±å¤éåå¸ã調æ´ããä»å ¥å¹æã®æ¨å®ãããæ£ç¢ºã«è¡ããã®ã§ãã
å¾åã¹ã³ã¢ãããã³ã°ã®æ義
å¾åã¹ã³ã¢ãããã³ã°ã®ç®çã¯ãä»å ¥ç¾¤ã¨å¯¾ç §ç¾¤ã®å ±å¤éåå¸ã調åãããå æå¹æããã¤ã¢ã¹ãªãæ¨å®ãããã¨ã§ããããã«ã¤ãã¦Rubinï¼2001ï¼ã¯æ¬¡ã®ããã«è¿°ã¹ã¦ãã¾ãï¼
"Matching on the propensity score can yield unbiased estimates of treatment effects if the propensity score model is correctly specified." ï¼å¾åã¹ã³ã¢ã«åºã¥ããããã³ã°ã¯ãå¾åã¹ã³ã¢ã¢ãã«ãæ£ããæå®ããã¦ããå ´åãä»å ¥å¹æã®ãã¤ã¢ã¹ã®ãªãæ¨å®å¤ãæä¾ãããï¼
å¾åã¹ã³ã¢ãããã³ã°ã®ããã»ã¹ã¯æ¬¡ã®ããã«é²ã¿ã¾ãï¼
-
å¾åã¹ã³ã¢ã¢ãã«ã®æ§ç¯ï¼ä¾ï¼ãã¸ã¹ãã£ãã¯å帰ï¼ã
-
å¾åã¹ã³ã¢ã«åºã¥ããããã³ã°ï¼ä¾ï¼1対1ãããã³ã°ãã«ãªãã¼ãããã³ã°ï¼ã
-
ãããã³ã°å¾ã®ãã©ã³ã¹ç¢ºèªã
-
ãããã³ã°ãã¼ã¿ã«åºã¥ãå æå¹æã®æ¨å®ã
ãã®ä¸æ ¸ã«ããã®ããå¾åã¹ã³ã¢ã®ãã©ã³ã·ã³ã°ã¹ã³ã¢ã¨ãã¦ã®æ§è³ªã§ãããã®æ§è³ªã«ãããå ±å¤éåå¸ãè¿ãåä½å士ããã¢ãªã³ã°ã§ããçµæã¨ãã¦æ£ç¢ºãªæ½å¨çµæã®æ¨å®ãå¯è½ã«ãªãã¾ãã
ãã¨ãã°ãå»çç 究ã«ããã¦æ°è¬ã®å¹æãè©ä¾¡ããéãå¹´é½¢ãæ§å¥ãåºç¤ç¾æ£ã¨ãã£ãå ±å¤éãä»å ¥ç¾¤ã¨å¯¾ç §ç¾¤ã§ç°ãªããã¨ãããã¾ããå¾åã¹ã³ã¢ãç¨ãã¦ãããã®å ±å¤éã調æ´ãããã¨ã§ãæ°è¬ã®å¹æãããå³å¯ã«æ¤è¨¼ãããã¨ãå¯è½ã«ãªãã¾ãã
å¾åã¹ã³ã¢ã®å°åºã«ããããã¸ã¹ãã£ãã¯å帰ã®å½¹å²
å¾åã¹ã³ã¢ã¢ãã«ãæ§ç¯ããéã«ããã¸ã¹ãã£ãã¯å帰ãããç¨ããããã®ã¯ã以ä¸ã®çç±ããã§ãã
-
åç´æ§ã¨è§£éã®å®¹æãï¼ãã¸ã¹ãã£ãã¯å帰ã¯ã¢ãã«ãæ¯è¼çåç´ã§ãå¾åã¹ã³ã¢ãæ示çã«è¨ç®ãããã¨ãã§ãã¾ããã¾ããå¤æ°ã®å½±é¿ã解éããããç¹ãç 究è ã«æ¯æããã¦ãã¾ãã
-
ãã©ã³ã·ã³ã°ããããã£ã®ä¿è¨¼ï¼ãã¸ã¹ãã£ãã¯å帰ãç¨ããã¨ããã©ã³ã·ã³ã°ããããã£ãé©åã«æºããå¾åã¹ã³ã¢ãå°åºããããã¨ãå¤ãã§ããé«ç²¾åº¦ã追æ±ããéç·å½¢ã¢ãã«ï¼ä¾ï¼å¾é ãã¼ã¹ãã£ã³ã°ï¼ã§ã¯ãå ±å¤éåå¸ãéå°ã«èª¿æ´ããããªã¹ã¯ãããã¾ãã
-
æ¨è«ã®ä¿¡é ¼æ§ï¼ãã¸ã¹ãã£ãã¯å帰ã¯ãçµ±è¨çæ¨è«ã«å¿ è¦ãªç¹æ§ï¼ä¾ï¼åå¸ä»®å®ãç·å½¢æ§ï¼ãæºããã¦ãããããå ææ¨è«ã®æèã§å®å ¨ã«å©ç¨ã§ãã¾ãã
ããã«å¯¾ããå¾é ãã¼ã¹ãã£ã³ã°ãªã©ã®é«ç²¾åº¦ãªã¢ãã«ãç¨ããå ´åãäºæ¸¬ç²¾åº¦ãåä¸ããä¸æ¹ã§ã以ä¸ã®èª²é¡ãçãã¾ãã
-
éå¦ç¿ã®ãªã¹ã¯ï¼è¤éãªã¢ãã«ã¯éå¦ç¿ã«é¥ããããããã©ã³ã·ã³ã°ããããã£ãæãªãããå¯è½æ§ãããã¾ãã
-
ãã©ãã¯ããã¯ã¹æ§ï¼ã¢ãã«ãè¤éã§ãããããå¾åã¹ã³ã¢ãã©ã®ããã«è¨ç®ãããã®ã解éãå°é£ã«ãªãã¾ãã
-
å¤æ¿ã®ãªã¹ã¯ï¼ä¸é¨ã®å ±å¤éç¯å²ã§æ¥µç«¯ãªå¤ãçæããå¯è½æ§ããããéã¿ä»ãææ³ï¼IPWãªã©ï¼ã§ä¸å®å®æ§ãå¼ãèµ·ããã¾ãã
ãããã£ã¦ãå¾åã¹ã³ã¢ã®ç®çããå ±å¤éåå¸ã®è¦ç´å¤ã¨ãã¦ã®ãã©ã³ã·ã³ã°ã¹ã³ã¢ãã§ããç¹ãèæ ®ããã¨ããã¸ã¹ãã£ãã¯å帰ã®å©ç¨ãé©ãã¦ããå ´åãå¤ãã¨ããã¾ãã
ããã«ãå¾åã¹ã³ã¢ã¢ãã«ã®é¸æã¯ãåã«äºæ¸¬ç²¾åº¦ã追æ±ããã®ã§ã¯ãªããç®çã«é©åããçµæãå¾ããã¨ãéè¦ããå¿ è¦ãããã¾ããå ææ¨è«ã®è¦³ç¹ã§ã¯ã解éå¯è½æ§ã¨é©åãªãã©ã³ã¹ã®ç¢ºä¿ãæåªå äºé ã¨ãªãã®ã§ãã
æå¾ã«
ã¨ããããã§ãå¾åã¹ã³ã¢ãå°åºããã¢ãã«ã¯ç¢ºçãæ£ç¢ºã«äºæ¸¬ã§ããå¾é ãã¼ã¹ãã£ã³ã°ç³»ã®ã¢ãã«ã®æ¹ãããã®ãããªè¨èª¬ãééã£ã¦ãããã¨ã¯ãç解ããã ããã§ããããï¼
å¾åã¹ã³ã¢ã®æ¬æ¥ã®ç®çã¯ããå ±å¤éåå¸ã®è¦ç´å¤ã¨ãã¦ã®ãã©ã³ã·ã³ã°ã¹ã³ã¢ãã¨ãããã¨ã常ã«é ã«å ¥ããããã§å¾åã¹ã³ã¢ãæ±ããã¨ããæ£ããå¾åã¹ã³ã¢ã®ä½¿ç¨ã«ã¤ãªããã¨æãã¾ãã
æå¾ã¾ã§ãèªã¿ããã ããããã¨ããããã¾ããã
markdownå½¢å¼ã§æ¸ããªãã¦è¯ãã¨ããã®ã¯é常ã«æ¥½ã§ãã...