Hello Saadia,

I think the answer depends on: (a) how volatile the variables are; and (b) what time intervals you are using.

If the variables tend to be stable over time, it may not make any difference as regards how they inter-relate. However, if there's been some explicit (treatment condition) or accidental intervention or event (for example, survivors of a tsunami who have lost their homes) that has occurred between data collection times, then I would treat the data sets independently.

If you really are unsure, then I'd consider CFA to test whether structures are the same across the data collection occasions.

Good luck with your work!

Hi Saadia,

Adding to the excellent point by David where it is important to consider whether you have introduced intervention or natural phenomenon occurring between time points.

Another important step is:

You need to run ANOVA or appropriate hypothesis tests to check whether there is significant differences between the measured variables at different time points. The results you got will then provide evidence to justify to combine or not to combine the data. Hope this helps.

Hi Saaida,

to keep things manageable, I would start with fitting separate models and then progress towards the full model. In this later stage, the option will be to test for time-invariant factor loadings (which is an essential assumption for stating the meaning equivalence of the construct). Likewise, you should investigate stability of the factor covariances.

Make sure to estimate cross-time error covariances of the errors of each item with itself one wave later. If it turns out that some items work at different occasions that does not mean that you are doomed. All you need is only two items with invariant factor loadings. The rest can differ with the extreme, that some items are part of the model in T1 and are omitted in later times (whereas others are added). This scenario needs some special treatment though (see the Baumgartner/Steenkamp paper).

I would further suggest, really testing the models with the chi-square test as the tests of invariance your not really convincing or reasonable when you omit the test in the baseline condition (that is that you do not the baseline model but suddenly do it when testing for differences).

Best,

Holger

P.S. here is some references

Baumgartner, H., & Steenkamp, J.-B. E. M. (1998). Multi-group latent variable models for varying numbers of items and factors with cross-national and longitudinal applications. Marketing Letters, 9(1), 21-35.

Golembiewski, R. T., Billingsley, K., & Yeager, S. (1976). Measuring change and persistence in human affairs: Types of change generated by OD designs. Journal of Applied Behavioral Science, 12, 133-157.

Millsap, R. E., & Hartog, S. B. (1988). Alpha, beta, and gamma change in evaluation research. Journal of Applied Psychology, 73, 564-574.

Schaubroeck, J., & Green, S. G. (1989). Confirmatory factor analytic procedures for assessing change during organizational entry. Journal of Applied Psychology, 74, 892-900.

Schaubroeck, J., & Green, S. G. (1989). Confirmatory factor analytic procedures for assessing change during organizational entry. Journal of Applied Psychology, 74, 892-900.

Thank you all of the learned researchers. i have kept all the points in my mind. as soon as i complete my data set and run the model i will get back again :) looking forward for all of yours guidance again in future.