1
$\begingroup$

What does it means if no CI was given for binary logistic regression analysis in SPSS output?

enter image description here

$\endgroup$
2
  • 3
    $\begingroup$ If you made a screenshot of SPSS output, post it on imgur and give us the link or we can insert the picture in your post for you. It would also be better to register your account. $\endgroup$
    – chl
    Dec 19, 2012 at 10:45
  • 2
    $\begingroup$ I would suggest you probably have linear separation of your data. The "signature" of this is large coefficients followed by extreme standard errors. You have categorical variables, so this usually means that one of the categories has either $0\text{%}$ or $100\text{%}$ response. $\endgroup$ Dec 26, 2012 at 6:16

2 Answers 2

3
$\begingroup$

I expect that you have discovered the Hauck-Donner effect in logistic regression where the likelihood flattens out and gives a much too large estimate of the standard error using Wald techniques. You need to find a tool that will give results based on likelihood ratios rather than Wald techniques when this happens.

$\endgroup$
1
$\begingroup$

It looks like that the standard errors for your coefficients (B) are so large (look at the column named S.E.) that the values of $\exp(B + SE)$ are practically infinite. You probably need a larger sample size or more relevant variables. Basically you couldn't reject the null hypothesis (no effect) with your model.

$\endgroup$
2
  • 1
    $\begingroup$ I also would suggest to investigate the existing data base for anomalies... $\endgroup$ Dec 26, 2012 at 8:08
  • $\begingroup$ In addition to what @user765195 wrote about overall sample size, you probably need to watch out for specific categories of predictors that have sample sizes in the single digits. $\endgroup$
    – rolando2
    Dec 28, 2012 at 0:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.