PDA

View Full Version : Independent t-test (Levene's test)

syunysi
04-25-2011, 08:37 AM
Hi everyone, I have a question about the Levene's test in the Independent t-test. I am comparing two measurements of both sexes, but the sample size of one of the sex is N=1, so the Levene's test does not show any figure (as shown as the image). My question is in this case is the the "sig (2-tailed) 0.126" still valid? Thank you so much in advance.

mathgeek10
04-25-2011, 11:19 PM
You should not run a t-test with a group that has one observation. There is an assumption with a t-test (and any parametric statistic) that your data approximates a normal distribution (bell curve). If you have only one data point there is no mean or variance to create any kind of distribution. Unfortunatley you other sample size is only 2 and that is also too small to run an inferential statistic on. The best thing for you to do with a sample size that small is to simply list each subjects scores in a chart. You will not be able to say statistically that any of the differences are meanful, but you can talk about trends.

tiff

zerostin
04-29-2011, 02:38 PM
I have a question about this. Let's say you have 2 variables with both 50 observations. A T-test will calculate the mean, and then check if those two means are different, right?
Let's say you calculate those two means by yourself. So, you have 2 mean values from 2 variables of each 50 variables. A t-test is than not possible, because N=1, but N is actually bigger. You just want to use those 2 means as input for the T-test instead of all the observation. Why is this not possible to do?

Dason
04-29-2011, 02:50 PM
The situation where you calculate the means yourself is fine because you actually do have replication. The original situation isn't fine because you don't. If you also told the computer what your sample standard deviations were along with the sample sizes it could run a t-test for you...

zerostin
04-29-2011, 02:54 PM
That's cool. I thought it wasn't possible, because every statistical program says that my N=1.
Is there some kind of tutorial how I can do that in SPSS? It's a long time ago when I worked with SPSS.
I don't see where I can input the SD. This is some example data

Variable | Obs Mean Std. Dev.
-------------+----------------------------------------------
gross_perf~e | 5603 .0389955 .2972835

vs

Variable | Obs Mean Std. Dev.
-------------+---------------------------------------------
gross_perf~e | 5159 -.017816 .1898832

Dason
04-29-2011, 03:16 PM
I don't know how to do it with SPSS. But it's not that hard to do by hand either.

zerostin
04-29-2011, 03:21 PM
Yeah, I should be able to do it by hand either. At least, I have done it before. However, I need to do this like 250 times, so it's a lot of work :P
I will try to figure it out somehow :)

Dason
04-29-2011, 04:12 PM
In that case you definitely want the computer to do it for you. Like I said I'm not familiar with SPSS but it can probably handle it. If you're doing it that many times are you planning on controlling the type I error through a multiple comparisons adjustment.

zerostin
04-29-2011, 04:30 PM
Yeah, I figured out a quick way to do it in Stata. Thanks for your help :)