Dear Jim,

Unequal sized sample groups are common in research. Due to random selection of samples, non response by samples, it is very likely to happen.

Problems that unequal sample size may lead to are-

1. Unequal variance. And thus conducting tests that assume equal variance will not be possible to run. A Levene's test may first be done to test if uneual sample in 2 groups are leading to unequal variance. If there is both unequal sample and unequal variance, statistical power and type I error ll be affected (Rusticus & Lovato 2014)

2. Equal group size increase statistical power

3. Problem with confounding variables.

Though Keppel 1993 states that equal size groups are not mandatory and softwares does the adjustment.

Nevertheless, here are some tests to embrace in case of both unequal sample size and Unequal variance.

1. Dunnett's T3 and C- for pairwise comparison T3 for small sample and C for large sample

2. Welch's test for unequal variance, which is a modified student's t test to increase test power with unequal variance

Now for unequal sample size but equal variance, use follwing parametric post hoc tests

Bonferroni

Fisher's test

Bonferroni

For non parametric,

Dunn pairwise

Dunn control.

Best,

Praheli

Jim,

First ask yourself why/how this result occurred in your survey in the first place.

Hello Jim,

Don't ever throw data away! All other things equal, the batch with 80 responses is likely to be a better estimate of that school's population of seniors than will the batch with 40 responses. You have two groups, they're unequal in size; don't worry about anything beyond that. Worry instead about the intended analyses, and whether they will address your research question(s) that drove the data collection in the first place.

You didn't indicate what proportion of the complete set of grade 12 students you were able to recruit successfully. If the sizes of 80 and 40 are pretty close to the full set that was extant at the time, then you've likely gotten samples that will credibly represent the entire population at the two schools. If not, then you'll need to be more cautious in whatever conclusions you draw from your data. If the response rates were markedly different between the two schools, you should likely do a little digging to identify possible explanation(s).

Good luck with your work!

Hi Jim Schofield ! I suppose you have a reason to compare the the two cohorts? Even so, I agree with David Morse that there is no argument that you have to have equal N when comparing two samples. As David Morse suggests, you should of course set the numbers for each class into a perspective (e.g., attrition analysis), but you should not toss away data!!

Best

JK

Hello there everyone thank you for the useful and prompt replies.

I see the key messages are don't scrap the data...response rate was nearly 100% from each school and the main reason for non-participation would have been absence from school that day Dennis Mazur ..hence as credible as possible David Morse

Praheli Dhar Chowdhuri would you say these tests are suitable for a Masters-level study? You suggestions look useful indeed, but I have not heard of these, perhaps due to my level of study.

The 2 cohorts vary in terms of mono-cultural John-Kåre Vederhus and multi-cultural.

Any further suggestions? Again, thanks so much, so far.

Jim

Ok. Don't forget confounding factors when discussing results. There was a similar comparison between monocultural (rural) & multicultural classes (Oslo) in Norway. In their results the authors "forgot" to mention that the mono-cultural/rural classes had 15 pupils per teacher, while in the multicultural Oslo classes there were 30 students per teacher with poor basic language skills among some pupils!

Hence, a comparison was quite unfair.

Best

John Kåre