ANOVA or t-test???

May 20, 2010 9:33 PM

[Stats-filter] Is there any reason to use ANOVA rather than a t-test when there are only 2 groups and 1 dependent variable?

These are examples of the questions I have to answer:

1. Is there any difference between primary schools and non-primary schools (a composite of high schools + technical schools) in teachers reported level of staff conflict?

2. Is there a difference between high schools and technical schools in teacher's reported level of time demands?

To me, it seems straightforward that I would just use an independent-samples t-test. However, Since we have been learning about ANOVA I am SURE the lecturer wants us to use ANOVA. But there are no questions with with more than two groups or two levels of dependent variable.

I thought maybe it had something to do with whether the distributions of the DVs were normal or not, or if variances were unequal. But for both t-tests and ANOVA in my textbook it says that normal population distribution is an assumption of the test, as is equal population variance.

I'm confused because it just seems to me that in these circumstances a t-test or ANOVA test would be the same thing. But I think there must be something I'm missing.

Any help GREATLY appreciated.
beccy

posted by beccyjoe to Science & Nature (9 answers total)

Homework filter? It's a close call. I'll bite, anyway. An ANOVA will give the same answer as an independent t-test for a single factor. The advantage of ANOVA is that they provide more information on the partitioning of variance (within-group and between-group), and of course, they're more flexible in that you can include as many factors and interactions as you like, so it's probably a good thing to learn to use them.

Still, I wish stats classes would get away from this ANOVA crap and start introducing generalized linear models etc.
posted by Jimbob at 9:40 PM on May 20, 2010

T-test should be fine, but if you run an ANOVA, you'll get the same result.
posted by k8t at 9:43 PM on May 20, 2010

Response by poster: Hm, thanks Jimbob and K8t.

yeah okay, stats-homework-filter.

Okay, so it's all about the information you get on the details of the variance... so you might as well use ANOVA because it gives you more information. That seems striaghtforward...

Then would it *ever* be more apt to use a t-test?
posted by beccyjoe at 9:56 PM on May 20, 2010

Well,

when to use a t-test

and this ANOVA v. t-test
posted by k8t at 10:05 PM on May 20, 2010

Sounds like an exercise/homeworkfilter.

ANOVA to begin with, then a manual post hoc with student t to figure out which ones are the interesting ones.
posted by porpoise at 10:38 PM on May 20, 2010

This isn't homework filter. She's not asking you how to do the problem, she's asking for confirmation of what she thinks, but is confused because her instructor (and the book) probably didn't explain it properly.
posted by anniecat at 6:31 AM on May 21, 2010 [1 favorite]

With two groups, ANOVA and a t-test are equivalent (it can be shown mathematically) if the t-test is done assuming equal variances in the two groups. There are versions of the two sample t-test for unequal variance, see e.g. the wikipedia entry for the t-test.
posted by Bijector at 8:58 AM on May 21, 2010

OK, piggy-backing on the OP:

Whenever I feel like I need to use ANOVA, I find examples for 2-3 variables. I need the generalized form for more! But I'm always stymied trying to find the general pattern to follow to extend ANOVA to more variables.

Can anyone point me to exemplars, clear formulations of N-variables, or anything else to help me?
posted by IAmBroom at 12:27 PM on May 25, 2010

I can't point you to any clear examples because it's probably a bad idea, and you don't need to use ANOVA for what you're doing, you should be using something else - ie. generalised linear models and multi-model inference. A three-way ANOVA is about as far as you can push it without people laughing at your experimental design and statistical interpretation.
posted by Jimbob at 2:13 PM on May 25, 2010