0
$\begingroup$

I am performing a binary logistic regression. I have developed a simple model which I am testing using the SPSS application. This first determines the predictive ability of a baseline model without using the independent variables and then adds in the variables to eatablish if the "new" model increases the % prediction.

The baseline model prediction % is already at 94.7%. The model with variables added pushes this up to 98.6%.

This is great, but none of the preditive variables is significant which confuses me.

Can you offer an explanation for this lack of significance?

Improve this question
$\endgroup$
3
  • $\begingroup$ can you post the variable significance and how you are determining if they are significant. $\endgroup$
    – mike1886
    Jul 21, 2014 at 13:18
  • 3
    $\begingroup$ This can happen, especially when your dependent variable has a landslide on either 1 or 0. If you already have 95% of 1's in the variable, then just by guess all of them as 1 will get you 94.7% accuracy. The predictor may have served to up the % slightly, and yet it may not have predicted enough of the variability to be a statistically significant predictor. $\endgroup$
    – Penguin_Knight
    Jul 21, 2014 at 13:26
  • $\begingroup$ This can easily happen when there's near-multicollinearity - you have a great model but none of the variables adds much given the contribution of all other variables. It suggests you could probably omit at least one variable without harming your model (not that you necessarily should) $\endgroup$
    – Glen_b
    Jul 22, 2014 at 1:13

1 Answer 1

Reset to default
2
$\begingroup$

This can happen, especially when your dependent variable has a landslide on either 1 or 0. If you already have 95% of 1's in the variable, then just by guessing all of them as 1 will get you 94.7% accuracy.

In that sense, % accuracy is not necessary an indicator of the predictor's performance. The predictor may have served to up the % slightly, and yet it may not have predicted enough of the variability to be a statistically significant predictor.

Improve this answer
$\endgroup$
1
  • $\begingroup$ I was fearful that the already high predictive accuracy of the base model would be an issue. My study concerns an investigation into customers who stay/leave a particular company. There are 2210 customers; 2092 stay, 118 leave. I've selected a set of independent variables & am attempting to determine if any are predictive. I feel with such a small number any significant factors will be hard to identify. The customers are segmented so I'll examine each segment. If any have a higher % departure perhaps something significant will stand out. Any further advice would be welcome. Many thanks $\endgroup$
    – Johnhm
    Jul 21, 2014 at 16:54

Your Answer

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

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