Home > Statistics Every Writer Should Know > The Stats Board > Discusssion
paired t-test vs. ANOVA
Message posted by tom s on June 28, 2000 at 12:00 AM (ET)
I have before and after treatment data of unequal population size (all who started the study did not finish) and I cannot remember whether a paired t-test is appropriate. Systat offers an "estimate model" under ANOVA and I have no idea what that is.
Your help would be greatly appreciated
READERS RESPOND:
(In chronological order. Most recent at the bottom.)
Re: paired t-test vs. ANOVA
Message posted by Chad Allen on June 28, 2000 at 12:00 AM (ET)
A paired-samples t-test is appropriate in this instance. However, for any results to be statistically reliable, I would advise dropping any participants who did not finish the experiment. Also, remeber to compute either an eta square or omega square to see how much of the variability is due to treatment effect (instead of sample size). Good luck. Chad Allen
lewingca@tribe.nlu.edu
Re: paired t-test vs. ANOVA
Message posted by Yossi on July 9, 2000 at 12:00 AM (ET)
Since you have two scores for each subject -- before and after -- a paired t-test is in order. However, this means that those for whom you do not have a result at time 2 will be dropped from the analysis.A couple of notes:
1)You can conduct a 'conservative test' and assume that those who dropped out would have had the same score at time 2 as at time 1. Thus, you would replace thier missing values with their initial score and do a paired t-test for the whole sample. (In biostatistics this is usually done under the heading of an Intent to Treat (ITT) analysis, where last available data point is carried forward to all subsequent measurements.
2) If your differences are not normally distributed, or if your sample is small, you should use a nonparametric test instead of a paired t-test. The noparametric alternative to the paired t-test is the Wilcoxon Signed-Rank test.
By the way, ANOVA and t-tests are perfectly equivalent. That is, if you do an ANOVA or a t-test you will get the same result. However, this particular ANOVA is a bit complicated. It is actually a Repeated Measures ANOVA and I wouldn't recommend it just because the output is difficult to understand.
Your $5 contribution helps cover part the $500 annual cost of keeping this site online.