Multi group analysis of structural equation modeling

Nishad -

Different number of variables, or different number of data points for each group? 

Subject matter is generally important, but if you are looking at a different set of variables in each group, then your analysis is highly dependent upon your subject matter. 

Assuming continuous data, if you mean that you have 9 groups, and total sample size for all 9 is 50, and you have one variable, then comparing means will likely be rather uncertain.  (I think proportions would be even less certain.) However, if each group has at least a few data points, with hopefully good data quality, then you might discern some likely differences if the population standard deviations for groups are small.  You could estimate population standard deviation for each group, and with the sample size for each group, estimate the standard errors of the estimated means.  You could compare groups pair-wise using confidence intervals of the differences in the paired estimated means, to see which seem to have a real difference, but expect some spurious results with so many comparisons.  There are multi-sample hypothesis tests, but that may not tell you much, and hypothesis tests are notoriously misinterpreted.  In particular, a stand-alone p-value is not very meaningful.  You would need a meaningful alternative hypothesis and type II error probability analysis or similar sensitivity study.  Also, people often use some 'standard' threshold, such as 0.05, which is a bad idea, because it should depend upon sample size and variability.

Cheers - Jim

Note: With small sample sizes, confidence intervals may not be able to depend so much upon the Central Limit Theorem, so a bit more uncertainty is implied.

Dear James,

Indeed, thank you so much for your suggestion

I wish you a wonderful day !

Best Regards

Nishad.

Dear James,

Good Morning

I have 9 groups ,

Group 1 has 4 Means,

Group 2 has 4 means,

Group 3 has 16 Means,

Group 4 has 4 means

Group 5 has 5 means

Group 6 has 5 Means,

Group 7 has 3 Means,

Group 8 has 5 Means

Group 9 has 4 Means

I would like to measure effectiveness these of the groups towards group effectiveness of the organizations. Please drop your suggestions.

In advance thank you so much for the response.

Best Regards

Nishad.

Dear Nishad,

Create one AMOS file for all data you have collected.

Method 1:

I assume group effectiveness of organization is a latent factor.

I also assume groups are formed after exploratory factor analysis

Then,

You can divide the files into five groups and with help of structural equation modelling measure the impact on latent variable .

Each groups variables considered as observed variable and group name is considered as latent variable.

These may be treated as antecedent variables.

Then find impact of all these on latent variable organizational effectiveness.

Method 2;

Use Bayesian estimation using Markov chain Monte Carlo simulation.

Dear Prasad,

Thank you so much for your suggestion, if there is any problem occurred while doing statistics, please allow me to ask questions.

I wish you a wonderful Sunday !

Best Regards

Nishad

[email protected]