1

If you were testing the hypothesis that age would have an effect on your dv for men, but not for women, would you measure this by doing a multiple linear regression with sex and age as the predictors for your DV or by splitting the data into male/female scores and then doing a simple linear regression with age as your predictor for each?

Amber
  • 11
  • What's the null hypothesis here? Is it that age has no effect for both sexes? That age has the same effect for both sexes? Something else? Or are there really several underlying hypotheses you want to consider here? – Glen_b Oct 06 '21 at 23:39
  • @Glen_b I have to test for two hypothesis, the first that age will have a positive effect on the DV for men and the second is that it will have no effect for women. So I guess the null hypothesis would be that age has the same effect for both sexes? I'm very new to statistics so I'm not completely sure how to go about answering that kind of question. – Amber Oct 06 '21 at 23:47
  • What do you mean that you want to show that age has no impact on women? What of the impact is $0.00001$? That value, after all, is not zero. // Frank Harrell, a Vanderbilt statistics professor and high-reputation member here, often says that age tends to have a nonlinear relationship with response variables. Do you definitely just want to test for a linear effect? // Have you plotted your two variables, with the points for men and points for women in different colors? – Dave Oct 07 '21 at 02:48
  • @Dave rank Yes it doesn't have a linear relationship with the data we have been given either, but we have to preform the test and then explain why the model was a poor fit because the assumptions for the linear regression model were violated. With the problem set we have to work through, I don't have any more information on the hypothesis other than that, it just says: Use linear regression to test the following hypotheses:
    1. The rate of the DV is higher for younger males, compared to older males.
    2. For females, the rate of the DV does not vary depending on age.
     – Amber Oct 07 '21 at 02:57
  • @Dave also thankyou for letting me know about Frank Harrell, I will look into his work more it sounds like it could be useful – Amber Oct 07 '21 at 02:58
  • Do you see why it’s a problem to show that the DV is exactly the same for women of all ages? I know they just mean to show that the coefficient has a high p-value, but that phrasing bothers me immensely (not your fault the teacher wrote the question poorly). – Dave Oct 07 '21 at 03:00
  • @Dave yes I agree, I understand what you mean. I think they are trying to use simple terms because most people taking the class won't have any background with statistics but it would probably sound strange to people more experienced. – Amber Oct 07 '21 at 03:06
  • Getting back to your question, how would you approach two models vs one combined model? How would each be analyzed? – Dave Oct 07 '21 at 03:09
  • @Dave With the simple regression model I would split the data into two groups for men and women and do a simple regression of age and the dv. With the multiple regression I would fit a multiple regression with age & sex as the predictor variables. I'm using R to do the analysis, but I'm not sure how to decide which would be most suitable. My inclination was to do the combined model but I'm really unsure. – Amber Oct 07 '21 at 03:28
  • 1
    I don’t see a good way to do this without an age:sex interaction term. – Dave Oct 07 '21 at 03:32
  • Do you specifically want to test that the effect on women is zero or just that the effect on men differs from the effect on women? – Dave May 05 '22 at 21:03

0 Answers0