如何用stata做稳健回归大量的线性回归模型是基于最小二乘法实现的,但其仍存在一些局限性。比如说,样本点出现许多异常点时,传统的最小二
å¦ä½ç¨stataå稳å¥åå½
大éç线æ§åå½æ¨¡åæ¯åºäºæå°äºä¹æ³å®ç°çï¼ä½å ¶ä»åå¨ä¸äºå±éæ§ãæ¯å¦è¯´ï¼æ ·æ¬ç¹åºç°è®¸å¤å¼å¸¸ç¹æ¶ï¼ä¼ ç»çæå°äºä¹æ³å°ä¸åéç¨ï¼æ¤æ¶åå¯ä»¥ä½¿ç¨ç¨³å¥åå½ï¼robust regressionï¼ä»£æ¿æå°äºä¹æ³ã
æä½
ä¸é¢ç稳å¥åå½ä½¿ç¨çæ¯ç¯ç½ªæ°æ®ï¼è¯¥æ°æ®æ¥èªAlan AgrestiåBarbara Finlayçã社ä¼ç§å¦ç»è®¡æ¹æ³ããåéå æ¬ç¾å½åå·ç¼å·ï¼sidï¼ãå·åï¼stateï¼ãæ¯10ä¸äººç¯ç½ªæ¡ä»¶æ°éï¼crimeï¼ãçæ´»å¨è´«å°çº¿ä»¥ä¸äººå£çç¾åæ¯ï¼povertyï¼åå亲人å£ç¾åæ¯ï¼singleï¼çãæ们éæ©ä½¿ç¨è´«ç©·çååç¶åµæ¥é¢æµç¯ç½ªçã
è·åæ°æ®
use https://stats.idre.ucla.edu/stat/stata/dae/crime, clear
summarize crime poverty single
å¯¼å ¥æ°æ®ï¼å¹¶æè¿°å个åéçç»è®¡ç»æï¼è¾åºè¡¨æ ¼ä¸å å«æ ·æ¬å®¹éãå¹³åæ°ãæ åå·®ãæå°å¼åæ大å¼ã
OLSåå½
å¨ç¨³å¥åå½ä¹åï¼æ们å è¿è¡OLSåå½ï¼è¾åºç»æå¦ä¸ã
regress crime poverty single
æ ·æ¬ç¹åæ
é¦å æ们éè¿âlvr2plotâç»å¶æ®å·®æ æå¾ï¼éè¿è¯å«ç¦»ç¾¤ç¹åé«æ æå¼ç¹ï¼æ æç¹ï¼è¿èè¯å«å¼ºå½±åç¹ãåå¦åå¨æ æç¹çè¯ï¼è¦ç¡®å®åªäºæ¯bad leverage pointï¼å¯¹äºè¿äºç¦»ç¾¤ç¹æ们è¦è¯ä¼°å®å¯¹æå模åçå½±åã
lvr2plot, mlabel(state)
ç±å¾ä¸æ们å¯ä»¥çåºï¼dcãmsãflä¸ä¸ªç¹æ®å·®è¾å¤§æè æ æå¼æ¯è¾é«ãåºå è·ç¦»æ¯æ æå¼ä¸æ®å·®å¤§å°ç综åæåºï¼ä¸è¬èè¨ï¼åºå è·ç¦»å¤§äº1ï¼åå¯è®¤ä¸ºè¯¥æ ·æ¬ç¹ä¸ºå¼ºå½±åç¹ãæ¥ä¸æ¥æ们计ç®åç¹çåºå è·ç¦»ï¼Cookâs Distanceï¼ï¼å¹¶è¾åºç»æã
predict d1, cooksdclist state crime poverty single d1ifd1>4/51, noobs
ç±ç»æå¯ä»¥çåºï¼dcç¹åºå è·ç¦»å¤§äº1ï¼è¡¨ædcè¿ä¸æ ·æ¬ç¹å¯¹äºåå½ç»æä¼äº§çè¾å¼ºçå½±åï¼å¨ä¹åç稳å¥åå½ä¸æ们ä¼å¯¹dcç¹è¿è¡ç¹æ®å¤çã
æ¥ä¸æ¥æ们åææ°æ®çæ®å·®ã使ç¨rstandardè¿ä¸å½ä»¤ï¼å®è¡¨ç¤ºæ ååæ®å·®çç»å¯¹å¼ã
predict r1, rstandardgen absr1 = abs(r1)gsort -absr1clist state absr1in1/10, noobs
稳å¥åå½
æ们使ç¨ârregâå½ä»¤è¿è¡ç¨³å¥åå½ï¼å¹¶è¾åºç»æå¦ä¸ã
rreg crime poverty single, gen(weight)
对æ¯æå¼å§çOLSåå½ï¼æ们åç°ä¸¤è å·®å¼è¾å¤§ã并ä¸ç¨³å¥åå½ä¸çæ ·æ¬ç¹æ°éæ¯50ï¼OLSåå½ä¸ä¸º51ï¼è¿æ¯å 为ç»è¿åé¢çåæï¼ç±äºdcè¿ä¸å¼å¸¸å¼ç¹å¯¹åå½ç»æå½±åè¾å¼ºï¼å æ¤å¨ç¨³å¥åå½ä¸æ们å°å ¶èå»ãä¸é¢çæä½è¡¨æå¨ç¨³å¥åå½ä¸ï¼dcæ ·æ¬ç¹æå æé为é¶ã
clist state weightifstate =="dc", noobs
ä¸é¢çå½ä»¤å±ç¤ºäºå ¶ä»æéè¾å°çè§å¯å¼ï¼ä¸è¬èè¨ï¼æ®å·®è¾å¤§çè§å¯å¼æéè¾å°ï¼ä¾å¦æ们ä¹åæå°çmsç¹ãå¨OLSåå½ä¸ï¼æææ ·æ¬ç¹çæéé½æ¯1ï¼å æ¤ç¨³å¥åå½ä¸è¶å¤çæ ·æ¬ç¹æéæ¯1ï¼å ¶åå½ç»æä¸OLSç»æè¶ç¸è¿ã
sort weightclist sid state weight absr1 d1in1/10, noobs
æ们è¿å¯ä»¥éè¿ç»å¶ååçæ¹å¼å½¢è±¡å°å±ç°è¿ä¸å ³ç³»ãä¸å¾ä¸æ¨ªåæ 表示å亲çï¼çºµåæ 表示ç¯ç½ªçï¼æ¯ä¸ä¸ªåå表示ä¸ä¸ªæ ·æ¬ç¹ï¼åå¿ä¸ºè¯¥æ ·æ¬ç¹å¨åæ ä¸çä½ç½®ï¼ååç´å¾è¶å¤§ï¼è¡¨ç¤ºè¯¥æ ·æ¬ç¹æéè¶å¤§ã
twoway (scatter crime single [weight=weight], msymbol(oh))ifstate !="dc"
æå±
æ们å¨ç¨³å¥åå½åæä¹åï¼å¯ä»¥ä½¿ç¨è®¸å¤åç»ä¼°è®¡å½ä»¤ï¼æ¯å¦testãmarginçãä¸é¢çæä½æ¯æ们æ§å¶è´«å°çä¹åï¼å¨ä¸åçå亲çä¸é¢æµç¯ç½ªçãæ们åç°ï¼éçå亲ççæé«ï¼ç¯ç½ªçä¹ç¸åºå°ä¸åã
margins, at(single=(8(2)22)) vsquish
大éç线æ§åå½æ¨¡åæ¯åºäºæå°äºä¹æ³å®ç°çï¼ä½å ¶ä»åå¨ä¸äºå±éæ§ãæ¯å¦è¯´ï¼æ ·æ¬ç¹åºç°è®¸å¤å¼å¸¸ç¹æ¶ï¼ä¼ ç»çæå°äºä¹æ³å°ä¸åéç¨ï¼æ¤æ¶åå¯ä»¥ä½¿ç¨ç¨³å¥åå½ï¼robust regressionï¼ä»£æ¿æå°äºä¹æ³ã
æä½
ä¸é¢ç稳å¥åå½ä½¿ç¨çæ¯ç¯ç½ªæ°æ®ï¼è¯¥æ°æ®æ¥èªAlan AgrestiåBarbara Finlayçã社ä¼ç§å¦ç»è®¡æ¹æ³ããåéå æ¬ç¾å½åå·ç¼å·ï¼sidï¼ãå·åï¼stateï¼ãæ¯10ä¸äººç¯ç½ªæ¡ä»¶æ°éï¼crimeï¼ãçæ´»å¨è´«å°çº¿ä»¥ä¸äººå£çç¾åæ¯ï¼povertyï¼åå亲人å£ç¾åæ¯ï¼singleï¼çãæ们éæ©ä½¿ç¨è´«ç©·çååç¶åµæ¥é¢æµç¯ç½ªçã
è·åæ°æ®
use https://stats.idre.ucla.edu/stat/stata/dae/crime, clear
summarize crime poverty single
å¯¼å ¥æ°æ®ï¼å¹¶æè¿°å个åéçç»è®¡ç»æï¼è¾åºè¡¨æ ¼ä¸å å«æ ·æ¬å®¹éãå¹³åæ°ãæ åå·®ãæå°å¼åæ大å¼ã
OLSåå½
å¨ç¨³å¥åå½ä¹åï¼æ们å è¿è¡OLSåå½ï¼è¾åºç»æå¦ä¸ã
regress crime poverty single
æ ·æ¬ç¹åæ
é¦å æ们éè¿âlvr2plotâç»å¶æ®å·®æ æå¾ï¼éè¿è¯å«ç¦»ç¾¤ç¹åé«æ æå¼ç¹ï¼æ æç¹ï¼è¿èè¯å«å¼ºå½±åç¹ãåå¦åå¨æ æç¹çè¯ï¼è¦ç¡®å®åªäºæ¯bad leverage pointï¼å¯¹äºè¿äºç¦»ç¾¤ç¹æ们è¦è¯ä¼°å®å¯¹æå模åçå½±åã
lvr2plot, mlabel(state)
ç±å¾ä¸æ们å¯ä»¥çåºï¼dcãmsãflä¸ä¸ªç¹æ®å·®è¾å¤§æè æ æå¼æ¯è¾é«ãåºå è·ç¦»æ¯æ æå¼ä¸æ®å·®å¤§å°ç综åæåºï¼ä¸è¬èè¨ï¼åºå è·ç¦»å¤§äº1ï¼åå¯è®¤ä¸ºè¯¥æ ·æ¬ç¹ä¸ºå¼ºå½±åç¹ãæ¥ä¸æ¥æ们计ç®åç¹çåºå è·ç¦»ï¼Cookâs Distanceï¼ï¼å¹¶è¾åºç»æã
predict d1, cooksdclist state crime poverty single d1ifd1>4/51, noobs
ç±ç»æå¯ä»¥çåºï¼dcç¹åºå è·ç¦»å¤§äº1ï¼è¡¨ædcè¿ä¸æ ·æ¬ç¹å¯¹äºåå½ç»æä¼äº§çè¾å¼ºçå½±åï¼å¨ä¹åç稳å¥åå½ä¸æ们ä¼å¯¹dcç¹è¿è¡ç¹æ®å¤çã
æ¥ä¸æ¥æ们åææ°æ®çæ®å·®ã使ç¨rstandardè¿ä¸å½ä»¤ï¼å®è¡¨ç¤ºæ ååæ®å·®çç»å¯¹å¼ã
predict r1, rstandardgen absr1 = abs(r1)gsort -absr1clist state absr1in1/10, noobs
稳å¥åå½
æ们使ç¨ârregâå½ä»¤è¿è¡ç¨³å¥åå½ï¼å¹¶è¾åºç»æå¦ä¸ã
rreg crime poverty single, gen(weight)
对æ¯æå¼å§çOLSåå½ï¼æ们åç°ä¸¤è å·®å¼è¾å¤§ã并ä¸ç¨³å¥åå½ä¸çæ ·æ¬ç¹æ°éæ¯50ï¼OLSåå½ä¸ä¸º51ï¼è¿æ¯å 为ç»è¿åé¢çåæï¼ç±äºdcè¿ä¸å¼å¸¸å¼ç¹å¯¹åå½ç»æå½±åè¾å¼ºï¼å æ¤å¨ç¨³å¥åå½ä¸æ们å°å ¶èå»ãä¸é¢çæä½è¡¨æå¨ç¨³å¥åå½ä¸ï¼dcæ ·æ¬ç¹æå æé为é¶ã
clist state weightifstate =="dc", noobs
ä¸é¢çå½ä»¤å±ç¤ºäºå ¶ä»æéè¾å°çè§å¯å¼ï¼ä¸è¬èè¨ï¼æ®å·®è¾å¤§çè§å¯å¼æéè¾å°ï¼ä¾å¦æ们ä¹åæå°çmsç¹ãå¨OLSåå½ä¸ï¼æææ ·æ¬ç¹çæéé½æ¯1ï¼å æ¤ç¨³å¥åå½ä¸è¶å¤çæ ·æ¬ç¹æéæ¯1ï¼å ¶åå½ç»æä¸OLSç»æè¶ç¸è¿ã
sort weightclist sid state weight absr1 d1in1/10, noobs
æ们è¿å¯ä»¥éè¿ç»å¶ååçæ¹å¼å½¢è±¡å°å±ç°è¿ä¸å ³ç³»ãä¸å¾ä¸æ¨ªåæ 表示å亲çï¼çºµåæ 表示ç¯ç½ªçï¼æ¯ä¸ä¸ªåå表示ä¸ä¸ªæ ·æ¬ç¹ï¼åå¿ä¸ºè¯¥æ ·æ¬ç¹å¨åæ ä¸çä½ç½®ï¼ååç´å¾è¶å¤§ï¼è¡¨ç¤ºè¯¥æ ·æ¬ç¹æéè¶å¤§ã
twoway (scatter crime single [weight=weight], msymbol(oh))ifstate !="dc"
æå±
æ们å¨ç¨³å¥åå½åæä¹åï¼å¯ä»¥ä½¿ç¨è®¸å¤åç»ä¼°è®¡å½ä»¤ï¼æ¯å¦testãmarginçãä¸é¢çæä½æ¯æ们æ§å¶è´«å°çä¹åï¼å¨ä¸åçå亲çä¸é¢æµç¯ç½ªçãæ们åç°ï¼éçå亲ççæé«ï¼ç¯ç½ªçä¹ç¸åºå°ä¸åã
margins, at(single=(8(2)22)) vsquish
温馨提示:答案为网友推荐,仅供参考