Can I do a t-Test with Unequal Sample Sizes? - The Unbalanced Research Design
HTML-код
- Опубликовано: 15 июл 2024
- Can you do a t-test when one group size is much smaller than the other? You can…but it may be wrong. We will explore the implications of unequal sample sizes on a t-test (although this applies to an ANOVA or other multivariate test, as well. We will learn about when heterogeneity of variances is a problem (and when it is not), the implications for power analysis, and the new t-test that can address the problem of heteroscedasticity (Welch’s t-test). You will learn where to find Welch’s t-test in SPSS.
This video teaches the following commands and techniques:
Welch’s t-test
Unbalanced Research Design
Heteroscedasticity
Link to video about randomly selecting a sample from a population of scores in SPSS: • Randomly Select a Samp...
Link to a Google Drive folder with all of the files that I use in the videos including the Bear Handout and datasets. As I add new files, they will appear here, as well.
drive.google.com/drive/folder...
Thanks a lot for your video. It was very helpful!
Great video. Thank you So much. I was stuck with the same problem. Your Video helped me.
Thank you so much. This is super useful 🤝
Thank you very much sir, right time right video
Thank you so much! Great video! My sample actually respects the ratio you mentioned. Only, I would have loved to find the source to back up the sample size in my dissertation. :-)
Still, I am relieved after I watched your great video. Thanks again!
I love your style of storytelling.
Thanks...I love including stories as part of the learning.
you are the best, loved it
Thank you so much for the comment and for watching the videos.
Thank You
Thank you so much!
You're welcome!
Thank you, very clear and helpfull🌺🌺
Thank you so much!
Great!
thank you SOO MUCH!!!
You're welcome!
Hi Prof, I have two independent groups treated with two different estrus synchronization protocols. I have progesterone concentration measured for both groups repeatedly at three-time points. I wanted to compare the progesterone concentration between two protocols on time 1, time 2, and time 3 rather than analyzing horizontally within groups. Since I have a low sample size of 18, I thought to analyze using multiple Mann-Whitney U tests. Could you tell me if this is appropriate? Or are there other tests to do the same analysis? Your help will be highly grateful.
Very useful! I'll dare ask a question, even if the video is old. You mentioned doing "randomly selection" to create a subset from the big group that matches the smaller group. Could this be combined with bootstrapping?
Hello!
I have used independent t-tests to examine gender differences and age differences regarding self-esteem in two countries. The groups had an unequal number of individuals and I got the following results:
Country 1:
t(61) = 0.58, p = 0.56, Cohen's d = 0.20
- Women n = 53, M = 28.20, SD = 5.23
- Men n = 10; M = 29.30, SD = 6.43
Country 2:
t(63) = 0.33, p = 0.74, Cohen's d = 0.09
- Women n = 46, M = 20.22, SD = 5.53
- Men n = 19; M = 29.84, SD = 4.81 and
Country 1:
t(61) = 0.79, p = 0.44, Cohen's d = 0.20
- Younger young adults n = 25, M = 27.72, SD = 5.05
- Older young adults n = 38, M = 28.82, SD = 5.64
Country 2:
t(63) = 1.10, p = 0.27, Cohen's d = 0.28
- Younger young adults n = 29, M = 29.38, SD = 5.68
- Older young adults n = 36, M = 30.83, SD = 4.95
I want to get an answer to the following question: Since the means between the groups are almost equal regardless of the unequal number of participants in them, this may (among other things) also indicate that a more even number of participants in the groups is not a guarantee that the results (the non-significance) had changed? Is this true and if so why? Can you explain this so I can understand it?
Sincerely
Jasna
Hey, I had a question if you can please help. So I have 2 sample groups of people from different locations and measured the time it takes for them to complete a task. One group had 39 people from one location and the other had 56 people from a different location. I wanted to compare which location completes the task faster.
-Therefore I have two unequal sample size groups, therefore unequal homogeneity of variance
-But I also have a non normal distribution of data (very positively skewed/right skewed) within each sample group (more data points near the lower time values on the left, indicating less time taken/faster to complete the task)?
Can I still use the Welch's T test, as that's also one of its assumptions?
I'm reading online that in those cases it's better to do a non Parametric test like Mann Whitney, Other reports online show that you can violate the assumption of normal distribution somewhat and still do a Welch T Test. Is this true? Or would it be better to do a log transformation of the data and then conduct a Welch T Test?
With my dataset, SPSS shows with Welchs T test: a significant difference between each group but Mann Whitney U test shows a non-significant difference between each group. What would be an appropriate statistical test to do then in this case to see whether the difference between my two groups are actually statistically significant?
I would appreciate any help
big thx subscribed
Thank you for the comment and the sub...recently hit 50K and it was amazing to see how many people benefit from the videos
thanks!
No problem!
Sir this is informative and am on my dissertation writing. Pease help me with the source or references
Thank you so much for your videos! I have a study with two sample groups that are the size of 186 and 16 (yes, sixteen). Welch's t-test said that p =
Yeah, that is a vastly different sample size. If you are cross-tabulating (using chi-square instead of parametric test), it will be very underpowered. If you can report means, then you should have scale data, so you could choose a parametric test instead. if you have categorical data, then the mean and SD are meaningless. Assuming a t test, yes, report the M & SD but be very cautious in your interpretation. Good luck
Thank you so much for the video. I have a question regarding this unequal sample size. In my research, I want to use the independent t-test result to check a pretest score, so to ensure that there are no significant differences between the two groups before they participate in this study. However, due to the research design, one group has a sample size of 30, and another of 60 (because the second group is meant to be 30 pairs). If that's the case, would you suggest to use a one-way ANOVA to compare 3 groups (segment 30 pairs into 2 groups of 30 people)? Or would you suggest to randomly choose 30 out of 60, and then do an independent t-test? Thank you very much.
If you have three groups and a pre-post design, then you would want to do a mixed repeated measures ANOVA. Sample size should not be a problem, but be sure to check the ANOVA assumptions first, as long as those are not violated, you can trust the results
@@ResearchByDesign Thank you very much for your suggestion. I will consider that.
hi... what if the groups are unequal and distribution in both is not normal. In the second scenario, unequal groups and one is normally distributed, and another one is not.
Hi! It is more informative. It would be helpful if you could leave the link of your video on random selection from a comparison group.
Great idea. Here is the link, in case you need it. I am going to add this to a card in the video itself. Thank you for the suggestion. ruclips.net/video/_VneI8FmYJU/видео.html
Thank you for your useful video. I am wondering what if my sample sizes are equal but my cell sizes are not?
By "cell sizes" are you referring to the cells of a chi-square? That is usually when people use "cells". If so, then as long as the cell size is greater than 5, you can use chi-square no problem.
Thanks, very informative and I almost found the solution to my problem, but alas, my data is a repeated measure of the same population, part of the population disappeared already so I am left with unbalanced design. Is the welch's t-test still my option? best regards
Please, please please do not run an independent samples t test where you had designed a paired t test. You would be only doing what is similar to p hacking and would render any theories you have developed useless. If you have to start over to get new data - fine , but don’t compromise your research. If your theories proves to be unfounded the only ethical thing a statistician can do is report it and start over. Believe me this happens more often than those teaching stats would have you believe. Rare and significant findings are supposed to be just that.
Hello. A question:
And can I use T-student of equal variances if I already proved that the variances are equal with Levene's test?
With unequal sample sizes data.
Yes you can, because the assumption of homogeneity of variance (equal variances) is separate from the assumption of equal sample sizes. If variances are equal, the unequal sample sizes will not bias the test outcome as much.
@@ResearchByDesign Thanks
Thank you! You’re video relieved my anxiety to my research. Hahaha. Although can I ask some questions? What if I have 50 individuals for the 1st group and 245 on my 2nd group? Would you recommend that we just get 50 individuals from the 245 so that the group will be balanced? And if I did that, will it not affect the generalizability of my study? Or should I should just leave the numbers as they are (50 vs 245)?
If I wanted to measure the significance of the difference between the revenue per customer in two different stores, one being the control group and the other offering a 10% discount, can I use the T.TEST to measure the delta in revenue per customer considering that the sample sizes are different because during the experiment more people purchased in one store over the other and so became customers. But.... the sample sizes are large 1k vs 1.1k? I guess what am I am really curious to understand is if the tolerance for unequal sample sizes increases as the sample size increases.
You could use an independent samples t-test. It will account for the unequal sample sizes and the two separate stores: ruclips.net/video/6eqm7zdPc3U/видео.html
@@ResearchByDesign Thanks for reply! Appreciate you!
I tried to use unequal design in R and it would not allow the Students t test, but did allow a Welch’s t test. Just to test further - both JASP and JAMOVI would not run either the Students or Welch’s analysis with unequal design.
Ahh...thank you so much for checking. R uses Welch's by default, but I did not know that it disallows Student's t ... and the JASP and JAMOVI info is surprising. I will look into that...thank you!!
I am trying to calculate the sample size to look up for five alleles frequencies in my population and after searching I found this equation: Sample Size = [z2 * p(1-p)] / e2 ,but someone just suggested that it should be calculated using the hardy equation!!! that's why I am asking if it can be used for sample size calculation!
Wow...you have me stumped. Sounds like you are doing genetic research and I understand that there are some aspect of calculations (such as sample size) that differs from social/psychological research. I don't know nearly enough about your field to offer any good advice. I hope that you can find someone who knows that answer. Best!
what about normality?
In categorical variables, equal sample sizes is the equivalent to normal distribution in scale variables. I have a class lecture video on scale variable normality coming up. Thanks!