https://www.scribbr.com/statistics/statistical-tests/

Answer to Que1: This sounds like a strange approach to me. Why would you want to combine factors that have been identified as being different enough in a factor analysis. You could do it by combining factor levels such as when you have Factor 1 with levels A,B,C... and Factor 2 with levels 1,2,3... Your new variable would look like A1,B1...C3... However this approach would result in a major loss of information of the previous factor and will reduce the data to the nominal scale. Why not use approaches that deal with multiple dependent values?

Answer to Que2 and 3: You might want to consider Generalized linear models: https://www.ibm.com/support/knowledgecenter/en/SSLVMB_23.0.0/spss/advanced/idh_idd_genlin_typeofmodel.html

Hello Arpitha,

If you've derived three factors and used orthogonal rotation for a final solution, there's no point in trying to combine them further as they are uncorrelated (this is largely a point that Sven Greving was making in his reply). Only if you use an oblique rotation and find non-negligible correlations among factors would it make sense to try for a scheme (e.g., a second-order factor solution) to try to "combine" them.

Was there some specific reason that you didn't mention for why it was important that your factors be combined in the first place?

As for question #2, I'd suggest you have a look at ordinal regression as an analytic method.

Good luck with your work.

Dear all,

Thank you for your valuable suggestions.

I wanted to combine the 3 factors to 1 to test if 2 IDVs predicts this one factor that's the DV of my study.

Now I realise it wouldn't make any sense to combine the factors further. Which is one reason why I needed experts opinion like all of you pointed out. Thanks!

Now I have arrived to a point to test each one of the 3 individual factors (IDVs in ordinal scale) if it predicts 3 DVs which are of different scales of measurements.

Please further suggest, which regression model would be best to use.

One of my colleagues suggested me to use Hierarchical regression.

I awful also consider ordinal logistic regression, any further suggestions or recommendations?

I seem a little bit confuse here. If you had used factor analysis, then you have several items on each of the variables. If I am not wrong, I don't think the initial rating (either categorical or ordinal) matters. This is because the computation of the scores (after considering possible reverse-scored items) will amount to a continuous rating for the constructs. Then, you could apply any statistical tool that fit analysing the variables as it were. Regression and the likes.

Dear Abayomi,

Thank you very much for your suggestions.

I have a quick question, what should be a minimum sample size to conduct a regression analysis or a pathway analysis?

I have around 90 items in my questionnaire but I have selected only few for the factor analysis. I have got 3 distinguished factors having 5 items each and these 3 factors are my latent variables. If I have to further move with path analysis after CFA, would my sample size (57) become a constraint?

Hello Arpitha,

One guideline for path/SEM models is to have 10-20 cases per parameter being estimated in the model. Including the measurement model (items associated with each of factors), you have 18 important parameters (not counting error terms, of which you'd have one per item) to be estimated. That's asking a lot of 57 cases.

Here's a good first reference for further reading (with some excellent citations for follow-up): https://www.statisticssolutions.com/how-to-determine-the-appropriate-sample-size-for-structural-equation-modeling/

Here's an online calculator that you may find helpful (though simulation is the safest way to get a good fix on sample size requirements): https://www.danielsoper.com/statcalc/calculator.aspx?id=89

Good luck with your work.

For Q2 I would suggest ordinal logistic regression. Just Google the term and you can find lots about it. Best, David Booth

Thank you very much for all your suggestions. I'm pleased with your guided references and the links shared @David Morse.

Many thanks,

Arpitha

Arpitha Vasudevamurthy ; I think you do not have a problem as regards that, however, the sample is also conditioned to some other factors - population size, specification or peculiarity of the population and research area of interest with the population. I also think David Morse suggested readings should fix most of this.

Multiple regression would be appropriate as your dependent variable is ordinal/scale data.

I am confused as are others apparently. Given that this appears to be a class assignment why did you start with a factor analysis since that approach seems to render the questions difficult if not impossible to answer? Best, David Booth