Adding age/maturity as a covariate would not really help you to examine what the effect of age is on e.g. the relationship between linear speed and agility performance; terms like age * linear_speed would be needed, and then the interpretation of the coefficients of these terms could get tricky - e.g., one unit increase in age*linear_speed adds coefficient amount of agility performance, what does that mean in the real world?

What I would suggest is a mixed-effects model (see e.g. Pinheiro & Bates:

Mixed-Effects Models in S and S-PLUS). These models are intended for grouped data - in your case, the grouping is by age - and produce both a population level model and group-level models on one go. Then it is possible to see if, for example, the coefficient of linear speed is significantly different in different groups. A couple of R packages exist for fitting mixed-effects models, at least lme4 and nlme. Other statistical software probably also have the capacity to fit them, but I am not familiar with them.

As a side note, one way to see if there are differences between let's say young and old players is to divide the data to two based on that grouping and fit standard multiple regression models to the two data sets separately. This can be a good initial analysis step before fitting mixed-effects models.

This is very helpful. Thanks, Kasper Kansanen

Have a look at https://methods.sagepub.com/reference/the-sage-encyclopedia-of-communication-research-methods/i9222.xml

https://sciencing.com/advantages-disadvantages-multiple-regression-model-12070171.html

https://www.researchgate.net/post/Why_would_we_ever_run_ANCOVA_instead_of_multiple_regression

this can be helpful https://www.reneshbedre.com/blog/multiple-linear-regression.html

I agree with Kasper's answer except that I don't think a mixed model is required unless you have repeated measures across ages or some kind of nesting structure (teams/clubs/schools).

Dear all scholars, i am new in the field. I want to know is this quary is either descriptive or explanatory?

What research strategy should follow?

As i know research design guide this issues. But, i want to know for clearing my idea.

Md. Faisal-E-Alam This looks to me to be explanatory because the researcher wants to know what variables contribute to agility performance rather

trying to predict it. If it was predictive he would be looking at what variables give the best perediction error. For details see Shmueli's classic paper which is attached, Best, D. Booth

Thank you Prof. D. Booth sir. In that case, there will single group of respondents or two group of respondents in Restrospective study sir?

Kasper Kansanen - I am exploring your suggestion re. splitting the sample and running independent multiple regression models. In my sample I have ~80-90 p's, split roughly into 3 different age brackets (let's call them young, middle, old).

Am I right to think that as a rough rule of thumb for multiple regression, you should have approx 20 p's per independent variable? I think I came across this info when exploring this analysis.

In which case, would splitting the data set this way be problematic?

Therefore, would there be an option to include age as a moderator (this is probably a more fitting word than my use of covariate in the initial Q - more along the lines of what I meant)?

James H Dugdale Splitting the data set can be a problem, if you are trying to fit models that have more independent variables than the sample size in the subsets (that's when the model becomes underdetermined and there is no unique solution). I, myself, have not heard of that rule of thumb; but naturally more data is (almost) always better.

I would see model selection as a crucial part of the analysis in those subsets, for example finding combinations of independent variables that e.g. minimize the Akaike information criterion. This might produce completely different models for different groups, and hence could be interesting... Another option is to use a model that tries to do model selection as it is fitted, such as lasso regression or Bayesian regression with priors that force parameters towards zero.

I must also admit that I might have been wrong and too hastily dismissed the original idea of moderation. I saw grouped data and tried to hit it with my hammer. Sorry about that. I am not too familiar with the moderated regression approach - I don't know if there are some tricks of the trade - but I can talk generally. In the full data, moderate with age: this means adding age as an independent variable, and adding all of the interaction terms such as age * linear_speed to the model. If you are trying to model the moderation effect of age with all k independent variables, this would lead to a model with 2(k+1) parameters; might be best to add the interactions one by one, seeing if the coefficient is significantly different from 0, and removing the interaction if it is not; there probably is no moderation effect on that variable.