Effects Plot and Coefficient Question

[stataddict]

Hi All,

I am running logistic regression model using variable selection, variable transform with maximum normal and then again variables transform with optimal binning.

I am getting quite good roc index and miss-classification rate.

In the effects plot I have many variables in blue (i guess that means positive effect) and one variable with red column (negative effect).

but in the Analysis of Maximum Likelihood Estimates many variables are negative and only one is positive i.e. exactly the opposite of the effect plot.

How to explain this?

Thank you!

                                 Analysis of Maximum Likelihood Estimates

                                                              Standard          Wald

Parameter                                   DF    Estimate       Error    Chi-Square    Pr > ChiSq    Exp(Est)

OPT_Age       01:low-34.5                    1     -1.1347      0.3570         10.10        0.0015       0.322

OPT_BMI       01:low-28.540152               1     -0.9561      0.3507          7.43        0.0064       0.384

OPT_FG        01:low-99.5, MISSING           1     -0.3387      0.3248          1.09        0.2971       0.713

OPT_HDL       01:low-45.4                    1      0.9830      0.3946          6.21        0.0127       2.672

OPT_LDL       01:low-114.3, MISSING          1     -0.5588      0.4448          1.58        0.2091       0.572

OPT_Ratio_New 01:low-0.6322109, MISSING      1     -0.7628      0.3756          4.12        0.0423       0.466

OPT_TC        01:low-201.25, MISSING         1     -0.1096      0.5006          0.05        0.8267       0.896

OPT_Tryg      01:low-190.2, MISSING          1     -0.1636      0.3848          0.18        0.6706       0.849

[jwexler]

Hi I have a quick answer for you.  If you look at the legend below the chart, red is positive and blue is negative.  That matches your Parameter Estimates.

Thanks,

Jonathan

[stataddict]

Thank you Jonathan!

So, the moral is: always look to the legend 🙂

regards,

Gio