Hello Stata People,
I am struggling with understanding how to pick one Random Effects Regression Model over another (using negative binomial).
In OLS linear regression I would look at R2, however, my xtnbreg ..., re command does not give me this value. How can I compare which model is better? I have the feeling it might have something to do with Wald Chi2, but really, I don't know
for example: I use control variables with different time specifications, eg. going back 3 years vs going back 5 years. Everything is the same except that. Now I have the two ouputs, where can I see which model better predicts my dependent variable so that I can choose which control variables to include in the end?
Model 1 Command:
. xtnbreg CitFirst Year IntegrationB CEOB AcRDIntensity TaRelSize IndustryRel TS3 AcPat3 TaPat3 RelBase3 CitFAll3, re
Results Model 1:
Model 2 Command:
. xtnbreg CitFirst Year IntegrationB CEOB AcRDIntensity TaRelSize IndustryRel TS5 AcPat5 TaPat5 RelBase5 CitFAll5, re
Results Model 2:
So, how do I know which model is better? Thanks in advance!!!
I am struggling with understanding how to pick one Random Effects Regression Model over another (using negative binomial).
In OLS linear regression I would look at R2, however, my xtnbreg ..., re command does not give me this value. How can I compare which model is better? I have the feeling it might have something to do with Wald Chi2, but really, I don't know
for example: I use control variables with different time specifications, eg. going back 3 years vs going back 5 years. Everything is the same except that. Now I have the two ouputs, where can I see which model better predicts my dependent variable so that I can choose which control variables to include in the end?
Model 1 Command:
. xtnbreg CitFirst Year IntegrationB CEOB AcRDIntensity TaRelSize IndustryRel TS3 AcPat3 TaPat3 RelBase3 CitFAll3, re
Results Model 1:
HTML Code:
Random effects u_i ~ Beta Obs per group: min = 1 avg = 4.7 max = 5 Wald chi2(11) = 72.34 Log likelihood = -700.72416 Prob > chi2 = 0.0000 CitFirst Coef. Std. Err. z P>z [95% Conf. Interval] Year -.3591065 .0596958 -6.02 0.000 -.4761081 -.2421048 IntegrationB -.4588381 .2820413 -1.63 0.104 -1.011629 .0939528 CEOB -.0611872 .2353371 -0.26 0.795 -.5224395 .4000651 AcRDIntensity -.6525456 .2689885 -2.43 0.015 -1.179753 -.1253379 TaRelSize -.2714529 .4692806 -0.58 0.563 -1.191226 .6483201 IndustryRel -.0054288 .3405246 -0.02 0.987 -.6728447 .6619872 TS3 2.15263 .6623846 3.25 0.001 .8543796 3.45088 AcPat3 .0000837 .0000921 0.91 0.364 -.0000969 .0002642 TaPat3 .0039008 .0041908 0.93 0.352 -.0043131 .0121147 RelBase3 -.0415342 .142761 -0.29 0.771 -.3213407 .2382722 CitFAll3 .00287 .0070084 0.41 0.682 -.0108661 .0166061 _cons -.1567299 .4392308 -0.36 0.721 -1.017606 .7041467 /ln_r .10801 .1968741 -.2778561 .4938762 /ln_s -.1608182 .3354247 -.8182385 .4966021 r 1.114059 .2193293 .7574058 1.638656 s .8514468 .2855963 .4412082 1.643129 LR test vs. pooled: chibar2(01) = 124.46 Prob >= chibar2 = 0.000
Model 2 Command:
. xtnbreg CitFirst Year IntegrationB CEOB AcRDIntensity TaRelSize IndustryRel TS5 AcPat5 TaPat5 RelBase5 CitFAll5, re
Results Model 2:
HTML Code:
Random-effects negative binomial regression Number of obs = 620 Group variable: dealnumber Number of groups = 132 Random effects u_i ~ Beta Obs per group: min = 1 avg = 4.7 max = 5 Wald chi2(11) = 67.73 Log likelihood = -703.15233 Prob > chi2 = 0.0000 CitFirst Coef. Std. Err. z P>z [95% Conf. Interval] Year -.3577128 .058282 -6.14 0.000 -.4719433 -.2434823 IntegrationB -.4594767 .2560041 -1.79 0.073 -.9612355 .0422821 CEOB -.0456475 .2331215 -0.20 0.845 -.5025573 .4112622 AcRDIntensity -.5919459 .2426813 -2.44 0.015 -1.067592 -.1162993 TaRelSize -.2345956 .4647577 -0.50 0.614 -1.145504 .6763128 IndustryRel -.0626072 .3458162 -0.18 0.856 -.7403945 .61518 TS5 1.741602 .6018254 2.89 0.004 .5620463 2.921159 AcPat5 .0000725 .0000639 1.13 0.257 -.0000528 .0001979 TaPat5 .0023788 .0015033 1.58 0.114 -.0005677 .0053253 RelBase5 -.0371322 .0354005 -1.05 0.294 -.1065158 .0322514 CitFAll5 .0047013 .0067243 0.70 0.484 -.008478 .0178807 _cons -.1014699 .4486066 -0.23 0.821 -.9807227 .7777829 /ln_r .1122034 .1803263 -.2412296 .4656364 /ln_s -.1602047 .3000423 -.7482768 .4278673 r 1.11874 .2017383 .7856612 1.593028 s .8519693 .2556268 .4731813 1.533982 LR test vs. pooled: chibar2(01) = 124.74 Prob >= chibar2 = 0.000
So, how do I know which model is better? Thanks in advance!!!
Comment