Dealing with Unequal Variances and Sample Sizes - SPSS (part 1)

SPSS provides a correction to the t-test in cases where there are unequal variances. However, when one has unequal variances and unequal sample sizes, this correction is no longer accurate. In this video, I demonstrate a simple approach to dealing with the problem of unequal variances and sample sizes.

[@gunnil08]

I'd like to thank you millions of times! your videos helped me a lot!

[@ohnyx6015]

Thanks for the wonderful videos. I was wondering if these videos are ordered? Would be very helpful for beginners.

[@monomor99]

you the best

[@atg0812]

I've learned SPSS with your videos! You are great!!
I have run into an issue...
I am running 4 Mann-Whitney U tests and I have unequal samples,
test O (N1= 270, N2=143),
test M (N1=93, N2=52),
test E (N1= 86, N2=50),
test L (N1=91, N2=41)
Is it a problem if I ran it like that, or should I alter N1 to the size of N2? It might help, as then I can compare mean ranks better?!
Also, do you know any papers that suggest equal sample sizes with justifications? I will need to justify my choices in my thesis, so it would really help.
Thanks!

[@rachaelmcqueen8711]

Hi there. Just wondering - I have different sample sizes BUT equal variances? Is it ok to use a t-test in this case?

[@vijayshanigarapu3877]

I have 3 groups of sample MICRO SMALL and MEDIUM is it possible to run regression

[@jennisoo-heelee1563]

Can I find any reference about whether this re-sampled data are representing original data? or what algorithm this function is using for randomization? I need some justification to use this approach.. Thanks.

[@how2stats]

The data are certainly a sample of the original data. I'm not aware of textbook reference for this procedure. However, I would expect something like this should suffice for the typical reader: "The 'select cases' utility in SPSS was used to select a pseudo-random portion (N = ?) of the original sample to ensure equal sample sizes across groups."

[@jennisoo-heelee1563]

Thank you for your explanation!

[@kulloemmanueldaata487]

Thanks so much for the videos. Please i need help on this. i am analysis some data and this is what i want to do using spss. can you help with the procedure?
To measure family sense of belonging, family autonomy, family control and school sense of belonging; a 3- level variable would be created: low, medium, high and the high category used as the reference category. These categories would be derived by scoring each of the responses, summing the scores and assigning the results to each of the categories. For example, the respondent’s social capital indicator ‘school sense of belonging’ would be made up of 6 statements such as: I feel like I belong to this school, I feel lonely at school; I feel like an outsider in my school, other students seem to like me etc. There would be 5 possible response categories and these would be scored as follows: strongly agree (1); agree (2); disagree (3); strongly disagree (4) and don’t know (5). For the analysis, scores would be grouped as: low SSB; medium SSB and high SSB.

[@how2stats]

What is your hypothesis?

[@kulloemmanueldaata487]

Hypothesis
Adolescent’s School Sense of Belonging (SSB) Hypothesis:
(a): adolescent’s self-rated health is positively influenced by their school sense of belonging.
(b) adolescent’s self - reported overall life satisfaction is positively influenced by their school sense of belonging.
(c) adolescent’s self - rated happiness is positively influenced by their school sense of belonging.

[@bjarbj944]

What kind of test should I use if I have much larger samples in the tens of thousands?

[@how2stats]

Sample size shouldn't really dictate what analysis you would do, in the context of this video. If you employ the video's strategy, you would still create random samples of data for each group to equal the smallest N.

[@atikhomthienthong9310]

Thank you. I have two set of data on vocabulary items provided by two groups of students. The total numbers of vocabulary items are not equal. The provided words are divided into two groups: correct and incorrect.
Group 1 (n = 218) - correct = 48 / incorrect = 170
Group 2 (n = 352) - correct = 82 / incorrect = 270
I want to know whether the difference between the correct and incorrect answers across the two groups. Can I use a chi-square?

[@how2stats]

Yes, I show how to do that here: ruclips.net/video/9dIBsa92QHY/видео.html

[@atikhomthienthong9310]

@@how2stats Thank you. But I want to look at the difference between the relative percentages across the groups. Can you give me more advice?

[@TokenFun105]

can't you just look at the Effect Sizes (cohen's d?) and if moderate then expect that differences will still be significant if samples were equal?