Binomial Logistic Regression

[William Peck]

I did the mechanics of Binomial Logistic Regression and got output but need assistance in understanding the results.

The study is to look at factors that most affect college graduation. So look at gender, high school rank, minorityYN, military experience YN, college placement tests (SAT Verbal, Math), Admissions Board total points, etc. It's a 10-year study with 12,000+ cases

I used this tutorial which I thought was very helpful, https://statistics.laerd.com/spss-tutorials/binomial-logistic-regression-using-spss-statistics.php.

But I don't understand what it's telling me ... mainly, in the Classification Table it says Predicted % is 100% for Graduation.

Here's the model summary

Model Summary
Step -2 Log likelihood Cox & Snell R Square Nagelkerke R Square
1 7819.768a                           .029                .054

a Estimation terminated at iteration number 5 because parameter estimates changed by less than .001.

Here's the Classification Table

Classification Table a
        Observed                                    Predicted
                                                         Status Percentage Correct
                                                 Graduate Separate
Step 1 Status Graduate 9251            0                 100.0
                                Separate 1361            0                      .0
        Overall Percentage                           87.2
a The cut value is .500


Variables in the Equation
                                  B S.E. Wald       df Sig. Exp(B) 95% C.I.for EXP(B)
                                                                                                Lower Upper
Step 1a
        Age_Start   .007 .050 .020      1 .889 1.007 .913         1.111
        Gender(1)        .153 .072 4.505 1 .034 1.166 1.012 1.343
        Minority(1) .215 .065 10.942 1 .001 1.240 1.091 1.408
        Feeder        18.904 3 .000
        Feeder(1)        .628 .304 4.269 1 .039 1.875 1.033 3.403
        Feeder(2)        .049 .338 .021        1 .884 1.050 .542        2.036
        Feeder(3)        .318 .303 1.106 1 .293 1.375 .760        2.488
        MilSvc(1)       -.264 .184 2.057 1 .152 .768        .535        1.102
        HS_Rank       .000 .000 .000        1 .999 1.000 .999        1.001
        HighSATM       .000 .001 .180        1 .671 1.000 .999        1.001
        HighSATV       .001 .000 2.527 1 .112 1.001 1.000 1.002
        WPM               .000 .000 111.097 1 .000 1.000 1.000 1.000
        Constant   3.124 1.162 7.223 1 .007 22.728

a Variable(s) entered on step 1: Age_I_Day, Gender, Minority, Feeder, PriorSvc, HS_Rank, HighSATM, HighSATV, WPM.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Rich Ulrich]

Note that the Classification Table is an "ancillary" statistic. That is, it is not used

in the computation of the regression; it is provided to help describe the results.

The reason that it is useless, here, is because the 0.5 cutoff is useless here.

87% of your cases are in one group, so you might imagine that "87%" as a sort

of average could be the median of all predictions; but the discrimination provided

by the equation is not strong; therefore, all cases end up being predicted as

being somewhat close to "0.87"  and thus member of that group defined by

> 0.50.  [Or, the similar logic for "13%" and < 0.50.]

There are a lot of commentaries on selecting different cutoffs. In your case -

What cutoff illustrates your "best success"?

Logistic regression can put out the predicted probabilities, so you can obtain

them and plot them by group, or create a complete ROC curve, or use arbitrary

cutoffs to look at various tables. I'd probably start with the arbitrary cutoffs,

just to see what I have.

--

Rich Ulrich

---

From: SPSSX(r) Discussion <[hidden email]> on behalf of William Peck <[hidden email]>
Sent: Tuesday, March 5, 2019 3:25 PM
To: [hidden email]
Subject: Binomial Logistic Regression

 

I did the mechanics of Binomial Logistic Regression and got output but need assistance in understanding the results.

The study is to look at factors that most affect college graduation. So look at gender, high school rank, minorityYN, military experience YN, college placement tests (SAT Verbal, Math), Admissions Board total points, etc. It's a 10-year study with 12,000+ cases

I used this tutorial which I thought was very helpful, https://statistics.laerd.com/spss-tutorials/binomial-logistic-regression-using-spss-statistics.php.

But I don't understand what it's telling me ... mainly, in the Classification Table it says Predicted % is 100% for Graduation.

Here's the model summary

Model Summary                  
Step    -2 Log likelihood       Cox & Snell R Square    Nagelkerke R Square
1       7819.768a                                  .029                  .054

a Estimation terminated at iteration number 5 because parameter estimates changed by less than .001.                   

Here's the Classification Table

Classification Table a                                 
        Observed                                                    Predicted           
                                                         Status          Percentage Correct
                                                 Graduate        Separate       
Step 1  Status  Graduate        9251                0                     100.0
                                Separate 1361                0                          .0
        Overall Percentage                                                         87.2
a The cut value is .500                                


Variables in the Equation                                                                      
                                  B      S.E.    Wald       df   Sig.    Exp(B)  95% C.I.for EXP(B)     
                                                                                                Lower    Upper
Step 1a
        Age_Start        .007    .050    .020          1  .889    1.007   .913             1.111
        Gender(1)               .153     .072    4.505   1       .034    1.166   1.012   1.343
        Minority(1)     .215    .065    10.942  1       .001    1.240   1.091   1.408
        Feeder                          18.904   3       .000                   
        Feeder(1)               .628     .304    4.269   1       .039    1.875   1.033   3.403
        Feeder(2)               .049     .338    .021            1        .884    1.050   .542            2.036
        Feeder(3)               .318     .303    1.106   1       .293    1.375   .760            2.488
        MilSvc(1)       -.264   .184    2.057   1       .152    .768            .535             1.102
        HS_Rank        .000      .000    .000            1        .999    1.000   .999            1.001
        HighSATM               .000      .001    .180            1        .671    1.000   .999            1.001
        HighSATV               .001      .000    2.527   1       .112    1.001   1.000   1.002
        WPM                    .000      .000    111.097 1       .000    1.000   1.000   1.000
        Constant           3.124 1.162 7.223     1       .007    22.728         

a Variable(s) entered on step 1: Age_I_Day, Gender, Minority, Feeder, PriorSvc, HS_Rank, HighSATM, HighSATV, WPM.                                                                      

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD