could perhaps be that your table is too large or too many data points for your computer to handle. You might want to use Monte Carlo simulation instead, or do a more traditional Pearson chi-square test (kzbin.info/www/bejne/mmWykJqcod6db9E)

The residuals post-hoc looks at a single cell and if the value is significantly different from what would be expected, not if it is significantly different from another cell. You might be more interested in a column proportion post-hoc test. Unfortunately I haven't made a video on this yet. It is available in my stikpetR library: install.packages("devtools") devtools::install_github("stikpet/stikpetR") library(stikpetR) You can then run the function ph_column_proportion(field1, field2). I have the same function in my Python stikpetP library, for which the documentation is found at: peterstatistics.com/Packages/python-docs/other/poho_column_proportion.html Hope this helps.

@@stikpet thank you very much! It helps a lot!!

@@stikpet and I would like to ask if I have multiple categories in both variables (unlike yours only female and male), is Bonferroni adjustment enough? or is there any other correction/adjustment I would need to do. thank you very much!

Bonferroni adjust for the number of tests performed. The residuals are based on the cell, so a test for each cell, with more columns you use more cells so that increases, the principal stays the same if you use 2 or 200 columns, just a lot more cells and therefor more tests.

@@stikpet thank you!!

strange. Just tested it myself again in SPSS version 29 and this should still work. Hard to tell what the problem might be without being able to replicate the error. Perhaps you can send me the file?

Not sure what you mean, since there are no superscripts used in this video. Could you perhaps mention the timecode in the video you are seeing this?

it's the value to enter if someone for example skipped the question. The 999 is just my own choice, you can chose any value you like, or leave it blank. I explain it in this video: kzbin.info/www/bejne/aJSnpp2ifqmLncU , or on my site at peterstatistics.com/CrashCourse/SPSS/SPSS-2.html. Hope this helps.

Using a rule of thumb from Cohen anything above 0.8 would be considered large. Hope that helps.

If you use the chart builder, you can set the order of the X-axis. At about 2:20 in the video you can see the 'Sort by' option of the X-axis. If you set it to 'value' it will sort them alphabetical. You could also first automatic recode the category variable (Index1) using Transform - Automatic recode from the menu bar, then use this new recoded variable. This will make it possible to sort the X-axis in any order you want. Hope this helps.

sorry about that. It was indeed a problem with the videos I made in the beginning of my KZbin channel.

thanks

glad the video helped. Good luck with your study/research/project.

first a one-way ANOVA then a post-hoc test. See peterstatistics.com/CrashCourse/3-TwoVarUnpair/NomScale/NomScale3a.html Hope this helps.

@@stikpet Thank you so much for your reply!! I have a gene expression as a continues values (50,28,49...) and a parameter as a nominal (0-1). When I try to do one-way anova and post-hoc in SPSS. It says Warnings Post hoc tests are not performed for Hao1_Median because there are fewer than three groups. Any idea about this? :(

strange, I thought I replied already. Anyway, simplified the one-way ANOVA determines if any of the means in the different categories could be different from each other. If at least one differs from another one a post-hoc analysis could determine which ones are different. Now if the one-way ANOVA is not significant, there is no point to perform a post-hoc analysis, there is also no point if there are only two groups (since if there are differences it can only be between those two). It seems you have only two categories (0 or 1), so no need for a post-hoc analysis. Actually the one-way ANOVA is then the same as an independent samples t-test. See for more info on this: peterstatistics.com/CrashCourse/3-TwoVarUnpair/BinScale/BinScale-T-test(independent%20samples).html

Unfortunately I don't have a Mac so cannot check if this works or doesn't on a Mac. It's hard to troubleshoot if I can't replicate the issue.

The ols function have in their documentation "Model statistics are given in the order of args.", so it is not the one/two/multi-way anova but the number of arguments. For a one-way ANOVA you need two: one that specifies the groups, and one that specifies the scores.

hard to see what the problem might be, since in the video it works. Perhaps a minor typo?

Not sure what you mean, but might be simply to make the table wider/longer?

yes, if there are no empty cells count should also work.

the official answer is always: it depends on the field you work in and you should not use rules of thumb. I always find that a bit unsatisfying, since not everyone is an expert in the field or has seen enough situations to know about this. Many rules of thumb find their origin in Cohen (1988). Cohen has a formula to convert eta-squared to Cohen f, and has some rule of thumb for Cohen f. I've done the conversions already and posted the rule-of-thumb interpertation at peterstatistics.com/CrashCourse/3-TwoVarUnpair/NomScale/NomScale3c.html. Hope this helps.

you're welcome. Glad it helped.

I think it was version 27, should still work. For older versions it is more tricky kzbin.info/www/bejne/fWbKoXl3hrmeeMk

Not sure if seeing the actual p-values is possible with SPSS for this post-hoc analysis. I do have I think something on how to do this with R, Excel or Python if you are interested.

Most welcome 😊

thanks. Greetings from The Netherlands :-)

uhm, which table?

@@stikpet independent sample test table and independent sample effect size table

@@SonikaYadav-gy6pm as the names imply they show the first shows the results of the independent samples t-test (and the Levene test for variances) and the other the effect sizes calculated. Each column shows something else which might be of interest. In the first table usually the p-value (sig.) is the most important. You can find out more about the test on my website.

@@stikpet yes

Glad it was helpful!

Glad it was helpful!

The Kruskal-Wallis test is used with two or more groups, so yes. Although, when only having two groups it actually becomes a Mann-Whitney U test.

A + or - value occurs due to the t-value from the test. In turn this t-value is + or - based on the numerator in that calculation. This is usually average of before - average of after. This will be negative if the average of after is greater than the average of before.

for nominal-ordinal usually a Kruskal-Wallis test is recommended, but your nominal variable seems with 'positve-neutral-negative' more ordinal. On my website (PeterStatistics.com) you can see which test I recommend for which combination of variables, and links to my youtube video on how to perform those tests. on a side note I almost reported you since my phone cut off your message and first didn't see the part 'ysis? I have a nominal....' 😁

if you select all the output, then right-click and select 'Export' you get the option to save the output as a Word document. You can then convert this Word document to a pdf.

to me it was one of those little gems I picked up somewhere and now use it all the time. Glad to have inspired others now with it 🙂

@@stikpet Totally agree. Disregarded it at first, didn't catch what it was for. Then kept getting errors somewhere. Extremely grateful.

thanks for the tip. Never knew about the aggregate function. Quite interesting how you can also use it to ignore hidden rows. For anyone else interested the microsoft help page is actually helpful: support.microsoft.com/en-us/office/aggregate-function-43b9278e-6aa7-4f17-92b6-e19993fa26df

perhaps a Friedman test? kzbin.info/www/bejne/m4q8eaWcgLKGfrc

Glad I could help!

oof, the bivariate normal was already quite a lot of work to figure out, and unfortunately never dug deeper to the multivariate version (which you probably need). My email is simply my youtube channel name with gmail.com :-).

I'm not sure which version this was. The video is 7 years old, so I guess it was version 24 or 25. I'm currently using version 29 and this should still work. Also the documentation of version 27 mentions the method shown (www.ibm.com/docs/en/spss-statistics/27.0.0?topic=sets-defining-multiple-response).

Glad to hear it!

there so many :-( I also learned a lot while trying to make this video :-).

sorry for the late reply. I'm not sure if this is possible using this method. However, if you create a cross table, including the total you can 'convert' the table into a clustered bar chart which should than have the result you are looking for. A video on how to convert a table to bar-chart can be found at: kzbin.info/www/bejne/qoCbdIZoebx6iacsi=IYbTtDuu6SQ45u5B