Hello Negin,

Is your concern that of how to interpret model results that were not statistically significantly different from zero? The direct answer is, the model failed to identify systematic relationships for the set of IVs with the DV, given the sample data that were collected.

Why might this occur?

1. There is no systematic relationship between the IVs and your DV in the population;

2. There is a relationship, but your study had insufficient power to detect it/them, due to the sample size (and the magnitude of the relationship in the population);

3. Your indicators are not as technically strong (e.g., reliability, validity) as would be desired; or

4. Some combination of the above.

Apologies if I've misunderstood your query.

Good luck with your work.

Hi David,

Thank you for your response, it was indeed helpful.

I was mainly wondering about "Model Fitting Information" from SPSS when running an ordinal regression. My results in the main regression is statistically significant, but the pre-tests for model fitting does not all of them indicate a good model fit. How can I manage this in the best way?

I think this is due to common method bias... you can try this

https://youtu.be/Y7Le5Vb7_jg

Hello again Negin,

Perhaps if you could post the overall model results, being sure to include the fit information, someone could address interpretation more directly.

Good luck with your work.

David Morse has said it all! To add further, you may try the following two ways for troubleshooting, since technically, there is a close call whether to accept/reject the model fit measures.

(1) First of all, since the data collection has already been made, small sample size could be a factor in model fit issues at this stage. In SPSS (recent versions), try Bootstrapping option and see if it improves the results. Caution: this is not a get-around trick to mask a set of troublesome data, if it is in reality. This could actually reveal another insight and contribute greatly in future research and data collection and/or model specification. If singularity is being warned, avoid bootstrapping in SPSS.

On another note, if the outputs from asymptotic methods vary largely from output of bootstrapping, this might well indicate that there is a problem in the data set. It could be because of outliers or highly skewed distribution of variables in the data set. On the other hand, if the sample drawn is representative of the population, bootstrapping will help immensely, otherwise bias will be there. (In addition, also choose the option of Parallel Line test to check the violation of ordinal regression assumption).

(2) Try with alternative "Link function". The link function is a transformation of the cumulative probabilities that gives estimation of the model. The following five link function options are available in SPSS, and try to see (by clicking on the Ordinal Regression--->Option--->Link Function) and choose one of the following functions based on your situation. Changing your “link function” based on the data characteristics may sometimes significantly change your outcome. However, ensure that the conditions match with your data first, otherwise choosing a method to “match” the “desired” output may conceal important insights and learning from the data!

• Logit/Probit-- Typically used for evenly distributed categories. Logit is the default selection in SPSS.
• Complementary log-log.-- Typically used when higher categories are more probable.
• Negative log-log-- Typically used when lower categories are more probable.
• Cauchit option-- Typically used when the latent variable has many extreme values.

Hope it helps.