Raw effects

[Luke Brown]

Good morning users,
I am investigating of the role of the dummy variable "crisis" on the dummy variable "hequity" (having a stock equity or not). Before running the entire model, I run a regression just between these two variables and I obtained a negative result (as I was expecting), however when I run the whole model, the effect of "crisis" on equity is positive. So, I am struggling to interpret these results. What can I conclude? Is the overall effect negative or positive?

logistic hequity crisis, vce(cluster YY1)

Logistic regression                               Number of obs   =      32123
                                                  Wald chi2(1)    =      96.96
                                                  Prob > chi2     =     0.0000
Log pseudolikelihood = -21563.372                 Pseudo R2       =     0.0022

                                 (Std. Err. adjusted for 6555 clusters in YY1)
------------------------------------------------------------------------------
             |               Robust
     hequity | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      crisis |   .7968009    .018381    -9.85   0.000     .7615772    .8336538
       _cons |   1.719106   .0307208    30.32   0.000     1.659936    1.780384
------------------------------------------------------------------------------

The whole model is as follows:

 logistic hequity i.hhsex i.Age i.Educ i.Race logincome logwealth crisis, vce(cluster YY1)

Logistic regression                               Number of obs   =      28816
                                                  Wald chi2(15)   =    5352.10
                                                  Prob > chi2     =     0.0000
Log pseudolikelihood = -11875.058                 Pseudo R2       =     0.3686

                                            (Std. Err. adjusted for 6512 clusters in YY1)
-----------------------------------------------------------------------------------------
                        |               Robust
                hequity | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
------------------------+----------------------------------------------------------------
                2.hhsex |   1.138956   .0458002     3.24   0.001     1.052636    1.232355
                        |
                    Age |
                 31-40  |   .9345882   .0571704    -1.11   0.269     .8289929    1.053634
                 41-50  |   .8150167   .0497118    -3.35   0.001      .723182    .9185132
                 51-60  |    .684578   .0436603    -5.94   0.000     .6041377    .7757289
                 61-70  |   .4769618   .0324377   -10.89   0.000     .4174401    .5449706
                   >70  |   .3351899   .0237262   -15.44   0.000      .291769    .3850727
                        |
                   Educ |
           High school  |    2.81198   .4109125     7.08   0.000     2.111673    3.744533
       College diploma  |   4.182021   .6176435     9.69   0.000      3.13092    5.585992
    Bachelor or higher  |   6.420295    .952637    12.53   0.000     4.800147    8.587276
                        |
                   Race |
Black/African American  |    .683181   .0341789    -7.62   0.000     .6193711    .7535649
              Hispanic  |   .4400365   .0256221   -14.10   0.000     .3925776    .4932327
       Asian and other  |   .5637518   .0458579    -7.05   0.000     .4806707    .6611929
                        |
              logincome |   1.846072   .0542434    20.86   0.000     1.742761    1.955509
              logwealth |   1.543859   .0192995    34.74   0.000     1.506492    1.582153
                 crisis |   1.087386   .0352158     2.59   0.010     1.020509    1.158645
                  _cons |   4.17e-06   1.37e-06   -37.76   0.000     2.19e-06    7.94e-06
-----------------------------------------------------------------------------------------

Thank so much in advance for your great help

[Carlo Lazzaro]

Luke:
you're trying to compare two very different models: no wonder that you get very different (or even opposite) results.
Please note that in the first (simple) logistic regression model, the role of -crisis- was not adjusted for other predictors (as, by definition, simle regression has one predictor only). whereas, in your second (multiple) logistic regression model, the role of -crisis- is adjusted for the other predictors.

[Steve Samuels]

Over 10% of the observations in the one predictor model were dropped from the whole model because of missing values in the added predictors. So the samples differ also.