Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "springathens@gmail.com" <springathens@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: Big Differences of Significance in Marginal Effect Estimation using margeff |
Date | Wed, 18 May 2011 16:07:21 -0400 |
Hi I am running Poisson regression. The following is the result of Poisson regression with beta coefficients . poisson y x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 Iteration 0: log likelihood = -55.606422 Iteration 1: log likelihood = -55.422123 Iteration 2: log likelihood = -55.42153 Iteration 3: log likelihood = -55.42153 Poisson regression Number of obs = 55 LR chi2(11) = 66.47 Prob > chi2 = 0.0000 Log likelihood = -55.42153 Pseudo R2 = 0.3749 ------------------------------------------------------------------------------ y | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- x1 | .0893232 .1247445 0.72 0.474 -.1551715 .3338179 x2 | -.092306 .1944331 -0.47 0.635 -.4733877 .2887758 x3 | -.0031832 .2545593 -0.01 0.990 -.5021103 .4957439 x4 | .2289895 .2351569 0.97 0.330 -.2319095 .6898885 x5 | .2321194 .1386073 1.67 0.094 -.0395461 .5037848 x6 | -.052605 .0149259 -3.52 0.000 -.0818592 -.0233508 x7 | -.044577 .0101308 -4.40 0.000 -.0644329 -.024721 x8 | .9073838 .5661389 1.60 0.109 -.2022281 2.016996 x9 | 3.787379 1.812603 2.09 0.037 .2347426 7.340016 x10 | .018216 .3453795 0.05 0.958 -.6587154 .6951475 x11 | .1276374 .3505633 0.36 0.716 -.5594541 .8147288 _cons | -34.79901 15.58228 -2.23 0.026 -65.33972 -4.258305 ------------------------------------------------------------------------------ As you can see there are a few independent variables that are not significant at .1 level. The following is the result after mfx. . mfx Marginal effects after poisson y = Predicted number of events (predict) = .51368381 ------------------------------------------------------------------------------ variable | dy/dx Std. Err. z P>|z| [ 95% C.I. ] X ---------+-------------------------------------------------------------------- x1 | .0458839 .06442 0.71 0.476 -.080375 .172142 1.52727 x2 | -.0474161 .10046 -0.47 0.637 -.244321 .149489 .044021 x3 | -.0016352 .13076 -0.01 0.990 -.257914 .254643 2.98182 x4 | .1176282 .12433 0.95 0.344 -.126053 .361309 2.30909 x5 | .119236 .06652 1.79 0.073 -.011137 .249609 1.67273 x6 | -.0270223 .00895 -3.02 0.003 -.044562 -.009483 51.9964 x7 | -.0228985 .00574 -3.99 0.000 -.034146 -.011651 59.5873 x8 | .4661084 .25856 1.80 0.071 -.040655 .972871 5.27121 x9 | 1.945515 .91814 2.12 0.034 .146001 3.74503 8.88313 x10 *| .009359 .17759 0.05 0.958 -.338702 .35742 .490909 x11 *| .0666069 .18605 0.36 0.720 -.298046 .43126 .381818 ------------------------------------------------------------------------------ (*) dy/dx is for discrete change of dummy variable from 0 to 1 The statistical significance did not change although marginal effects at mean and their corresponding standard errors changed. The following is the result of average partial effects using margeff . margeff Average partial effects after poisson y = E(y) (expected number of counts) ------------------------------------------------------------------------------ variable | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- x1 | .0926945 .0066891 13.86 0.000 .0795841 .1058048 x2 | -.0956625 .0103587 -9.23 0.000 -.1159653 -.0753598 x3 | -.003299 .0135354 -0.24 0.807 -.0298279 .02323 x4 | .2393958 .0129356 18.51 0.000 .2140425 .2647492 x5 | .2427261 .0077471 31.33 0.000 .2275421 .25791 x6 | -.0545182 .0008759 -62.25 0.000 -.0562348 -.0528015 x7 | -.0461981 .0006235 -74.10 0.000 -.0474201 -.0449761 x8 | .9403796 .0307734 30.56 0.000 .8800649 1.000694 x9 | 3.925102 .100002 39.25 0.000 3.729102 4.121102 x10 | .0190514 .0187024 1.02 0.308 -.0176047 .0557075 x11 | .1410915 .0211994 6.66 0.000 .0995414 .1826415 ------------------------------------------------------------------------------ As you can see, those insignificant variables turned out to be very statistically significant. Why does this happen? Thanks SR * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/