Low Wald chi-square values in GEE

[Christian Sprinkmeyer]

Hello everyone,

I am running a GEE -xtgee- model with an identity link function, a Gaussian (normal) distribution and AR(1) correlation. Furthermore, I use robust variance estimators to control for heteroscedasticity. In order to account for fixed effects I specified the panel with -xtset- based on company and year.

My problem is that I receive relatively small Wald chi-square values which are mostly non-significant. While this might certainly be an indicator that the model is simply not representative and not explaining the DV, I am wondering if there might be other reasons for these low Wald chi-square values. For example, are there specific issues with some of the variables included? Or are there any other ways to optimize the Wald chi-square value?

Especially given the fact that the number of observations as well as the number of variables included are relatively large, the low Wald chi-square values surprise me. Any info or help on this issue would therefore be really appreciated.

GEE population-averaged model                   Number of obs     =      1,125
Group and time vars:           gvkey fyear      Number of groups  =        161
Link:                             identity      Obs per group:
Family:                           Gaussian                    min =          2
Correlation:                         AR(1)                    avg =        7.0
                                                              max =          9
                                                Wald chi2(21)     =      25.71
Scale parameter:                  1.600558      Prob > chi2       =     0.2178

                                               (Std. Err. adjusted for clustering on gvkey)
-------------------------------------------------------------------------------------------
                          |               Robust
                       EO |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------------------+----------------------------------------------------------------
           Median_TMT_Age |  -.0783608   .0455389    -1.72   0.0853    -.1676154    .0108937
                 TMT_Size |    .033364   .0317819     1.05   0.2938    -.0289274    .0956555
                 Firm_Age |   .0357644   .0809936     0.44   0.6588    -.1229801     .194509
                Firm_Size |   .1577051   .0768751     2.05   0.0402     .0070328    .3083775
          Financial_Slack |   .0733506   .0434145     1.69   0.0911    -.0117403    .1584414
         Past_Performance |   .0175702    .034472     0.51   0.6103    -.0499936     .085134
   Environmental_Dynamism |   .0145041   .0679373     0.21   0.8309    -.1186506    .1476587
    Competitive_Intensity |    .033028   .0658973     0.50   0.6162    -.0961284    .1621844
                          |
                    fyear |
                    2007  |  -.1792236   .1806779    -0.99   0.3212    -.5333458    .1748986
                    2008  |  -.2379827   .1927414    -1.23   0.2169    -.6157488    .1397835
                    2009  |   -.353077   .2056169    -1.72   0.0860    -.7560788    .0499249
                    2010  |  -.3551309    .213963    -1.66   0.0970    -.7744908    .0642289
                    2011  |  -.2732624   .2210608    -1.24   0.2164    -.7065336    .1600088
                    2012  |  -.1856549   .2277244    -0.82   0.4149    -.6319866    .2606767
                    2013  |  -.1176107   .2245817    -0.52   0.6005    -.5577827    .3225612
                    2014  |  -.1420228   .2292282    -0.62   0.5355    -.5913018    .3072561
                          |
                    sic_1 |
            Construction  |          0  (omitted)
           Manufacturing  |  -.2285387   .2568444    -0.89   0.3736    -.7319444    .2748671
Transportation/Utilities  |  -.0719184   .3129316    -0.23   0.8182     -.685253    .5414163
        Retail/Wholesale  |  -.2857057   .3145684    -0.91   0.3637    -.9022484    .3308371
                 Finance  |   .1264767   .2891284     0.44   0.6618    -.4402046     .693158
                Services  |  -.0735055    .311783    -0.24   0.8136    -.6845888    .5375779
                          |
                    _cons |   .3444924   .3007342     1.15   0.2520    -.2449357    .9339205
-------------------------------------------------------------------------------------------

Thanks in advance.

Christian

[Marcos Almeida]

Hello Christian,

Welcome to the Statalist/ Stata Forum.

Actuall, the non-significant p-value for the "omnibus" Wald test does not indicate that the coefficients are "mostly non-significan", but none of them (but the constant) is supposed to be statistically different from zero.

And that is what is really happening.

Without mentioning the rationale and the way variables were measured (precision and accuracy included), I would suggest to start with a model without autoregressive option. Also, considering the time variable was "clusterized" with the group variable, I would consider excluding "fyear" from the list of predictors, just as a start up.