Selecting parameters from different models

Hello everybody, I'm totally new to statistics and it is the point where I have to use statistical analysis for a work I'm currently struggling with. I run the regression analysis for 25 different data files (for same independent and dependent variables) to come up with equations for each data file. So I have 25 equations. Now I'm wondering, what could be the best analysis to select one equation for all of 25 data files? Thank you,

[John BG]

that sounds like a lot of data. Can you hang a sample and draft a crashing start script so interested readers can fix it?

[Star Strider]

What’s different about the individual data files? Different experimental conditions? Different populations?

By ‘equations’ do you mean different regression models, or simply different estimated parameters for the same regression model?

[amberly hadden]

Thank you Star and Jhon, Each data file is the result of different experiment and yes different population. Yes, different regression models. I want to pick one regression model which could be representative of the entire data set. In each models (e.g., y = ax+b) values of a and b are changing while exponents of regressor are same due to physical definition between parameters. I want to come out with one model with one value of a and b which could be best represntative of entire data set.

[amberly hadden]

Thank you star, may be student T test could be of some use. But I appriciate your help.

[Star Strider]

Please do not use the t-test alone to compare your groups!

I would use multcompare or its friends. Pairwise tests such as the unpaired t-test (in this instance) need to be adjusted for multiple comparisons. The multcompare test will do that for you. Consider the 'scheffe' option for 'CType'.

The analysis-of-variance (ANOVA) would be the option I would choose to compare each characteristic between groups. In this instance you are looking for no significant differences between them.

[amberly hadden]

Sounds good, Thank you so much star

[Star Strider]

My pleasure.

Answer 1

[Star Strider]

‘Each data file is the result of different experiment and yes different population.’

Then don’t combine them. One size does not fit all. Report them individually, and discuss the differences.

[amberly hadden]

Thank you, start. But it is not possible as I have 140 data sets. and displaying 140 equations is not practically acceptable. Also variation between a and b is not huge (eg., range is between 300 - 500 )

[Star Strider]

If you have 140 different experiments with 140 different populations, combining them may not be statistically possible unless you can demonstrate that the subjects in the experiments were by all relevant characteristics not statistically significantly different. (Using the multcompare function on each characteristic could be an option.) If they weren’t different, you could possibly combine all your data in one file and do the regression on that.