0
$\begingroup$

I want to do an analysis where my model contains 1 dependent variable which is continuous (eg. SCORE of respondents) and around 10 independent variables (eg- Age, income(continuous), Gender, disease status(binary), level of education, no. of working hours(categorical)). Will it be appropriate to use ANOVA?

Or should I use factorial ANOVA(i.e. Two-way ANOVA)? If none of them is appropriate, which method would be apt for this kind of model?

$\endgroup$
4
  • 1
    $\begingroup$ No, neither factorial anova nor two-way anova are appropriate. You need to use general linear models (aka multiple regression with factors). Note BTW that "generalized linear model" and "general linear model" are not synonyms. You want the former. The latter is something more advanced that I doubt you want to deal with. $\endgroup$ Dec 8, 2017 at 5:07
  • 1
    $\begingroup$ I've removed the "generalized linear model" tag and replaced it with "multiple regression". $\endgroup$ Dec 8, 2017 at 5:08
  • $\begingroup$ @GordonSmyth , I want some more clarification based on your response. 1. Why is two-way ANOVA inappropriate? Is it because it will involve a lot of interaction terms? Or is it something else? 2. From my understanding, the Generalized linear model will be utilizing multiple regression techniques while a general linear model is a bivariate analysis.Please correct me if I am wrong here. $\endgroup$
    – Aurora
    Dec 9, 2017 at 18:03
  • $\begingroup$ 1. Two-way anova is for 2 factors. You have 10, so two-way anova isn't even a possibility. 2. Your understanding of GLMs (ized and otherwise) is entirely wrong. Please a google search so you can at least get in the right ball park and start asking more meaningful questions. $\endgroup$ Dec 10, 2017 at 3:07

1 Answer 1

0
$\begingroup$

First of all, what is the purpose of your analysis is important.

If you want to evaluate the difference in scores (y) between the status of the disease (binary; 0 or 1), I recommend multiple linear regression in order to adjustment for confounding factors(age, income, gender, ...).

It is possible to estimate the difference in score between groups (disease status) adjusted for other covariates.

$\endgroup$
1
  • $\begingroup$ To J-H Yoon's point, I don't have a main (exposure) effect which I am keen on analyzing. I was thinking of multiple regression too. But wouldn't that imply 'adjusting' to one main exposure effect instead of how each of the exposure effect in the model gets predicted? $\endgroup$
    – Aurora
    Dec 9, 2017 at 18:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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