1
$\begingroup$

I used a one-way MANOVA to study the effect of age groups on the average of height and weight and found that they were significant. Then I used a two-way MANOVA to study the effect of age groups and gender on the averages of height and weight and found that gender had an overall effect but not age groups.

Is there any logic in the result obtained using the above two methods?

Update. The data are of 75 males and 75 females and I randomly made age groups 18-25, 25-30, 30-35, 35-40 and 40+. The number of members in each group varies.

$\endgroup$
3
  • $\begingroup$ Do you have the same number of subjects in different groups (particular age + particular gender), or substantially different? $\endgroup$
    – amoeba
    Commented Jul 25, 2014 at 13:09
  • $\begingroup$ The data is of 75 males and 75 females and I randomly made age groups 18-25, 25-30, 30-35, 35-40 and 40+. The number of members in each group varies. $\endgroup$
    – user52672
    Commented Jul 25, 2014 at 14:52
  • $\begingroup$ This might be the reason. Imagine for example that males are higher than females, but you have a lot more young males than females. Then there will be no effect of age when you put both age and gender in the model (which is correct), but if you only look at age, then it will seem as if age makes a difference: young people are higher. This would be wrong, as there are simply more young males (in the sample), and they are higher. $\endgroup$
    – amoeba
    Commented Jul 25, 2014 at 15:53

2 Answers 2

1
$\begingroup$

Categorizing a continuous variable is not recommended (e.g., MacCallum et al., 2002), as @JoelW pointed out. Although I am not aware of any articles investigating the effects of cetegorization on MANOVA, it would be safe to stay away from it.

As @amoeba suggests, you should plot all variables using something like the pairs function in R. If distributions of the variables look reasonable, I would ran a MANCOVA model with age as a continuous variable along with gender.

$\endgroup$
0
$\begingroup$

What ages are you using? If you are dealing with pre-adults, it may be that gender and age are confounded, with girls growing big at younger ages than boys.

$\endgroup$
2
  • $\begingroup$ Age ranges from 18 - 55. So, I just randomly made groups like 18-25, 25-30, 30-35, 35-40 and 40+. Please do suggest an alternative method to study whether height and weight differ across age groups. $\endgroup$
    – user52672
    Commented Jul 25, 2014 at 14:53
  • $\begingroup$ Since categorizing age loses information, and since you have only 2 research quesions, you might calculate two correlations (age with height, and age with weight), perhaps using a smaller alpha (to keep the experiment-wise alpha at .05, or whatever value you are using). Also, you might plot your data to see if it seems to be linear. $\endgroup$
    – Joel W.
    Commented Jul 25, 2014 at 17:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.