Age adjusted scores

[Emma Menden]

I am currently trying to adjust a set of scores for age. I have both raw scores on a scale from 0-100 as well as percentile ranked scores. I have read that using the regress function followed by predict can help with this but am not unable to generate results which make sense. Is the regress function correct to use? For example

regress overall_reading_score age
predict

Many thanks for the advice in anticipation.

[Carlo Lazzaro]

Emma:
welcome to this forum.
If your dependent variable is continuous, you can actually use -regress- to explain variation in the dependent variable due to a 1-year increase in your unique predictor (be sure that you do not have to plug in more predictors to get a reasonable model).
With the following code:

regress overall_reading_score c.age

you are using the -fvvarlist- notation (see -help fvvarlist-) and are asking Stata to consider -age- as a continuous predictor.
With this other setup:

regress regress overall_reading_score c.age##c.age

you are using -fvvarlist- to include both a linear and a squared term for -age-. This approach can be useful if you suspect turning points (that is, the dependent variable reaches a maximum [minimum] that correspond to a given age included in the -age- range and then decreases [increases]).

A less advisable approach is considering -age- as a categorical variable (with as many levels as the years included in -age-):

regress overall_reading_score i.age

-fvvarlist- notation takes again the scene but this time asks Stata to consider -age- as categorical-.

-predict- (please, see -regress postestimation-) will simply allow you to see what happens when you sum up the constant of your regression model to the coefficient for -age- multiplied by the -age- of a given observation included in your dataset.

[Emma Menden]

Hi Carlo,
Thank you so much for responding as I am quite new to this! I am currently working on some analysis which looks at the impact of initial low performance on reading scores over 4 time frames. The full sample I have to work with is 1300 however there is substantial variation in age so I need to address this as a confounder when looking at scores. I am trying to generate a set of age adjusted scores which can allow me to work with the whole sample and not limit it. I am therefore trying to get age adjusted scores for each unique observation. Am I on the right track with this if I use this code:
regress overall_reading_score c.age predict age_adjusted_score Apologies if I'm misinterpreting this!
Best wishes,
Emma

[Carlo Lazzaro]

Emma:
if you're dealing with repeated measures of reading score on the same sample, you should consider panel data regression ( see help xtreg- and related entry in Stata .pdf manual);
-other issue: have you got other predictors in addition to -age-? I do not think that you can have an informative model with -age- only in the right-hand side of your regression equation (by the way, gender can be a case in point).
- not quite. The correct code would be:

regress overall_reading_score c.age##c.age
predict
list

Try it out and see what happens.