If you are using ANOVA, then use the option to include age as a covariate. If you are using regression, then enter it as another independent variable.

Use age as a covariate in the statistical analysis. remove any age effects.

Hi Chloe,

what statistical method do you use for your main analysis? In general, a common procedure is to add age as additional predictor or covariate into your analysis and see whether the results change.

It depends on the type of analysis, but the recommendations above are reasonable. If you are doing correlation, choose a partial correlation instead. Without the info of your analysis all of this is a shot in the dark.

It all depends on your method of analysis but in general, using age as a covariate will get the job done.

I think you need to remove the outliers age from your data set then try again. 

There isn't a single correct answer and you'd need to provide more information about the analysis. There are situations where adding age as a covariate is wasted effort (repeated measures anova with no age x factor interactions), situations where it can be dangerous (adding it as covariate in a repeated measures model without centering the covariate) and situations where there are better options (such as adding it as an offset).

I don't understand the characterization of using age covariates as "dangerous." Are you talking about possible mis-specification of the model without adding additional interaction effects? If so, I would worry about whether one had theory for such effects, or whether the recommendation is to test every possible effect in the model?

Also, I fail to see the danger in not centering a covariate. Since the correlation between a variable that is centered and one that is not centered is 1.0, this should not produce anything more than a algebraic respecification of any model in Ordinary Least Square.

As for Repeated Measure Anova, it is a rather primitive approach and, definitely more likely to produce misspecification compared to using lagged regression. I I had to choose something that was "dangerous" that would be it.

I adjust for age differences in spatial cancer studies. I am looking for any unusual cancer rates that are not associated with age.

David,

there is nothing inherently dangerous about adding age as a covariate (the point was that there are situations where it can be) or about repeated measures ANOVA (e.g., in a traditional experimental design). The original question didn't include sufficient information to provide a useful answer.

There is no danger in not centering a covariate in general. However, there are a number of software packages (notably SPSS) where including uncentred covariates does mess up the model because of the way the model is parameterised (in terms of difference scores) and fitted.

Delaney, H. D., & Maxwell, S. E. (1981). On using analysis of covariance in repeated measures designs. Multivariate Behavioral Research, 16, 105-123.

(also see attached link for a simple summary).

http://www.psyc.bbk.ac.uk/research/DNL/stats/Repeated_Measures_ANCOVA.html

I suggest that you use a three-way Chi-Square to hold age constant when compared to another important independent variable. For example, if gender (sex at birth) is correlated to voting yes or no on a controversial issue, holding gender constant makes sense to see if yes or no votes differ, even when gender is constant.  

But if the slopes (of the covariate and dependent variable) differ between the categories of your ANOVA, the ANCOVA will be invalid. You can't just chuck all confounding variables into the covariate list and ignore the assumptions of ANCOVA.