[Help] Logistic regression and goodness-of-fit

#1
Hello.

I have a question on "goodness-of-fit" of an ordinal logistic regression run that I did.

The results of the ordinal logistic regression were:
1) some factors were significant in the expected direction - this was good.
2) log of odds was OK
3) Log-likelihood was OK
4) concordant, discordant, Somers' D, Goodman-Kruskal Gamma, Kendall's Tau-a were poor but acceptable.

BUT, the goodness-of-fit was unacceptable.
- Pearson and Deviance chi-square test was below p=0.05.

In this case, what should I do?
Are the results totally invalid? (even if the factors that I was interested in were significant in expected direction?)


-----------------------------minitab--------------------

Ordinal Logistic Regression: story2 versus yearseq, quarter, ...

Link Function: Logit


Response Information

Variable Value Count
story2 1 70
2 77
3 57
Total 204


Logistic Regression Table

Odds 95% CI
Predictor Coef SE Coef Z P Ratio Lower Upper
Const(1) -3.31532 0.849168 -3.90 0.000
Const(2) -1.51931 0.822304 -1.85 0.065
yearseq 0.253472 0.122420 2.07 0.038 1.29 1.01 1.64
quarter 0.259496 0.137301 1.89 0.059 1.30 0.99 1.70
YuckFactor 0.416405 0.119275 3.49 0.000 1.52 1.20 1.92
publicopinion 0.228037 0.189416 1.20 0.229 1.26 0.87 1.82
ReportType
Editorial -0.805892 0.800065 -1.01 0.314 0.45 0.09 2.14
Feature -0.0803011 0.721771 -0.11 0.911 0.92 0.22 3.80
News -0.0092407 0.441279 -0.02 0.983 0.99 0.42 2.35
News Report -1.53155 1.50650 -1.02 0.309 0.22 0.01 4.14


Log-Likelihood = -206.758
Test that all slopes are zero: G = 31.634, DF = 8, P-Value = 0.000


Goodness-of-Fit Tests

Method Chi-Square DF P
Pearson 187.104 140 0.005
Deviance 192.493 140 0.002



Measures of Association:
(Between the Response Variable and Predicted Probabilities)

Pairs Number Percent Summary Measures
Concordant 8897 64.6 Somers' D 0.33
Discordant 4345 31.6 Goodman-Kruskal Gamma 0.34
Ties 527 3.8 Kendall's Tau-a 0.22
Total 13769 100.0
 
Last edited: