Reporting significant H&L results

#1
Hi, 4 years ago I did a logistic regression analysis as part of my PhD with a total sample of over 5,000. Now I am editing my thesis and I realise I forgot a lot about statistics and I am a bit lost. 3 out of my 18 models are significant at 0.004, 0.014 and 0.020 respectively. So this mean I should not draw any conclusions on these models right? should I avoid discuss them further than saying they were significant and not a good fit, and that they do not describe the data? I used the hosmere and Lemeshow test in SPSS. Thanks for reading
 

noetsi

Fortran must die
#3
Generally if a model is not significant you would not bring it up other than to say that the model, as compared to a variable, was not significant. How did you decide the model was not significant.

Goodness of fit is an entirely different issue than model signficance.
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
Did you eventually whittle down the models to one that you actually used or did you fit a **** ton of models and report on multiple? I don't know if I would get too worked up about H-L test. If is likely better to make decisions based on plotting the observed vs predicted or calibration plots than a single all or nothing test. Model preference can also be examined using AICC.