Dependent variable in % and GLM model

[marcelo ribeiro]

Hallo, I have a Dependent variable (Y), normaly distributed, that goes from 0 to 1. I checked the correlation between Y and my Independt variables (pwcorr Y X1 X2..., sig) and I found significance for some of them. However, when I do a model (glm Y X1 X2 X3.. , family(binomial) link(logit) robust) just one X is significant and the others no. I checked the correlation between the indep variables and some present 0,7 correlation others 0,3 or 0,1. Should I take them out of the model? how can I fix this model? which tests should I use?
Thank you

[Nick Cox]

We can't easily tell. If your response has mean near 0.5 and small variance then a linear probability model might well work fine in practice. If your mean is near 0 or 1 then a model should surely respect the range of the response and use a link function that ensures predictions within (0, 1).

Why not give a sample of your data or show a plot of your data, e..g a scatter plot matrix?

Best to say that responses between 0 and 1 are proportions, fractions or probabilities, not percents.

[marcelo ribeiro]

Good night Nick,
Thank you very much for your answer! the mean of my DV is .69 and with a std dev of .18, so in this case should I avoid a linear probability model? In my case which link function should I use?
I'm sending the sample of my data. I've been using the glm with the IVs but just "women in manager position" is significant, but doing separately some regression I found significance for some IVs and also significant correlation with most part of them...

Thank you

[Nick Cox]

A logit link sounds a good idea. I stop short of trying to read a spreadsheet file for reasons we do explain in https://www.statalist.org/forums/help#stata 12.5. As every new message prompt reminds you, posters are asked to read that first.

[marcelo ribeiro]

Hallo Nick, thanks for the link, I didn't know, I'm new here, my first time using the forum.
Anyway, I'll try to use the logit as you advise me. Thank you.

Just to post what I did in the right way (I hope): my DV is MenComp and ranges from (0,1) with mean 0,69. Some of my IVs ranges from (0,1) as well but others are continuous, should I transform all of them in (0,1) as well? I just have one variable significant, but all of them are significantly correlated to my DV. I guess I'm doing something wrong.

Sample:

input double(MenComp GlobalGenderGap WomenManagerPosition Adolescentbirthrate HumanCapitalIndexFemale)
.11 .68   32 45.9 .643
.14  .7 26.5 45.6 .682
.18   .    .    .    .
 .2  .69    0   46    .
.23   .    .    .    .
.33   . 27.2 27.3    .
.35 .76 46.3 13.5 .773
.37 .72 56.7 52.8 .558
.37 .68 20.7  4.6 .759
.38   .    .    .    .
end

Listed 10 out of 175 observations

Command used:

glm MenComp GlobalGenderGap WomenManagerPosition Adolescentbirthrate HumanCapitalIndexFemale, family(binomial) link(logit) robust

note: MenComp has noninteger values

Iteration 0:   log pseudolikelihood = -33.035298 
Iteration 1:   log pseudolikelihood = -33.018227 
Iteration 2:   log pseudolikelihood = -33.018218 
Iteration 3:   log pseudolikelihood = -33.018218 

Generalized linear models                          No. of obs= 76
Optimization     : ML                                    Residual df=71
                                                                    Scale parameter = 1
Deviance         =  12.06759915                   (1/df) Deviance =  .1699662
Pearson          =  10.92649627                    (1/df) Pearson  =  .1538943

Variance function: V(u) = u*(1-u/1)                [Binomial]
Link function    : g(u) = ln(u/(1-u))                  [Logit]

                                                                            AIC             =  1.000479
Log pseudolikelihood = -33.01821785                BIC             = -295.4145

-----------------------------------------------------------------------------------
                  |               Robust
          MenComp            |      Coef.       Std. Err.       z       P>|z|        [95% Conf. Interval]
------------------+----------------------------------------------------------------
  GlobalGenderGap       |   -2.38153   1.628387    -1.46    0.144    -5.573109    .8100495
WomenManagerPos~n |  -.0090151   .0063643    -1.42   0.157    -.0214888    .0034587
Adolescentbirth~e         |  -.0054699   .0035937    -1.52   0.128    -.0125135    .0015736
HumanCapitalInd~e      |   -1.78425   .8346752    -2.14    0.033    -3.420183   -.1483164
            _cons                 |   3.988976   1.026239     3.89   0.000     1.977585    6.000366
-----------------------------------------------------------------------------------

.

[marcelo ribeiro]

Sorry:

 glm MenComp GlobalGenderGap WomenManagerPosition Adolescentbirthrate HumanCapitalI
> ndexFemale, family(binomial) link(logit) robust
note: MenComp has noninteger values

Iteration 0:   log pseudolikelihood = -33.035298 
Iteration 1:   log pseudolikelihood = -33.018227 
Iteration 2:   log pseudolikelihood = -33.018218 
Iteration 3:   log pseudolikelihood = -33.018218 

Generalized linear models                          No. of obs      =        76
Optimization     : ML                              Residual df     =        71
                                                   Scale parameter =         1
Deviance         =  12.06759915                    (1/df) Deviance =  .1699662
Pearson          =  10.92649627                    (1/df) Pearson  =  .1538943

Variance function: V(u) = u*(1-u/1)                [Binomial]
Link function    : g(u) = ln(u/(1-u))              [Logit]

                                                   AIC             =  1.000479
Log pseudolikelihood = -33.01821785                BIC             = -295.4145

-----------------------------------------------------------------------------------
                  |               Robust
          MenComp |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
------------------+----------------------------------------------------------------
  GlobalGenderGap |   -2.38153   1.628387    -1.46   0.144    -5.573109    .8100495
WomenManagerPos~n |  -.0090151   .0063643    -1.42   0.157    -.0214888    .0034587
Adolescentbirth~e |  -.0054699   .0035937    -1.52   0.128    -.0125135    .0015736
HumanCapitalInd~e |   -1.78425   .8346752    -2.14   0.033    -3.420183   -.1483164
            _cons |   3.988976   1.026239     3.89   0.000     1.977585    6.000366
-----------------------------------------------------------------------------------

.