Thank You for the lecture. My question would be which test should I use if I want to see the p-value of the change comparing the same nominal data before intervention and after intervention (nominal variable has 4 options but respondent can choose just one of them both times - before and after intervention)?
@@petemiksza Thank You. I tried it, yet instead of p-value system shows a letter ,,a", which is described as ,,all test fields do not have exactly two values" below the table of hypothesis.
@@VaivaGegužinskaitė oh, I see, I’m sorry the Cochran Q test works for comparisons of a dichotomous outcome on more than two occasions - you may want to try the “marginal homogeneity test” : www.statisticssolutions.com/free-resources/directory-of-statistical-analyses/mcnemar-marginal-homogeneity-sign-wilcoxon-tests/
@@anacatarinasantos7752 In this video, I used an older version, in which you had to "double-click" on the summary output for some non-parametric tests to get more detailed information. In versions 27 and 28 the more detailed information that only appeared in previous versions when you "double clicked" the output are now presented by default - all the information should be there. In other words, there's nothing else to see.
Hi, may I know do u have any suggestions for me to do correlation? I had actually collected the agreement level on the benefits with 5 point likert scale. On the other hand, i do have data on their willingness to use with yes or no choice. May I know which method is suitable?
Hello - if you have one dichotomous variable representing independent groups responses - yes or no - and one ordinal variable (e.g., a rating scale item), then you can use a Mann-Whitney U test to compare the rating scale responses as a function of whether participants responded yes or no to the dichotomous item. If you can treat the ordinal categories of the rating scale item as nominal, then you could use the “Cramer’s V” coefficient. There’s also a specific coefficient that I do not know much about that is suited for these data called Freeman’s theta.
Thank you for this video!! ♥️♥️ I just want to confirm if I should use Chi-square in correlating nominal data with four categories to ordinal data with three categories? Some said that I should use point biserial correlation with this one.
Yes, you could use Chi square in that circumstance. In doing so, you would be treating your ordinal variable as a purely nominal variable. I do not believe a point-biserial correlation would be appropriate in this case. It is typically used when one variable is dichotomous (i.e., two categories) and the other variable is scalar (i.e., interval- or ratio-level).
Hello Thanks for the informative lecture. One question . I need your help and clarification in something. I am conducting a study to determine the difference in physical activity pattern before and after covid 19. If I want to compare before and after for multiple outcome variables ( never Exercised,once, twice, and three times) which statistical test is best to compare the difference before and after?
Hello, It sounds like you want to compare responses the same group of participants made to an item asking how often they exercised with response options of never, once, twice, or three times on multiple occasions (i.e., pre- and post-COVID-19). If you treat those data as "interval/ratio-level" measurements, then you could use a dependent samples t-test to look at whether the mean difference between the two time points is significant. If you treat those data as "ordinal-level" measurements, then you could use a Wilcoxon test to look at whether the difference between the responses at the two time points is significant. Given there are only 4 possible outcomes, you might consider treating those data as "ordinal," especially if you have a small N; even though-technically-the response options represent "equal intervals" (i.e., 0-1-2-3), the responses seem likely to be distributed in a fashion that is more similar to an ordinal measurement.
Thank you for your quick response! I appreciate it. Yes, I am comparing the resposes for the same group but before and after a certain period of time. I have more than 4 possible outcomes. The outcomes I have are: Never Once Twice Three time More than three times I have a large sample of 370 participatns. I wonder if this kind of data can be treated as ordinal data? Can it be? If yes, then I can go for wilxon sign rank test? Another related question. What if I also have other multiple categorical variables and I also want to test the difference before and after, such as the following: How many minutes did you exercise in a typical week before covid 19 pandemic? Less than 30 minutes 30 minutes 30 minutes - less than 1 hour 1 hour - less than 2 hours 2 hours And then the same question again "How many minutes did you exercise last week? " Which test can also be suitable for this type of data? Thank you very much
@@halahala3524 Hello, each of the response modes you described in your reply include "ordinal" categories (i.e., more than three times, less than 30 minutes) - so, I would definitely recommend using a non-parametric test for comparing changes in responses over time with those items - for two measurement occasions, the Wilcoxon test is appropriate.
Some argue that if a Likert scale variables involves enough response options (e.g., more than 5 scale points) and the distribution of the values roughly resembles a normal distribution, then you can apply parametric tests. This is generally the rationale many psychologists use. However, many statisticians would say it is never appropriate to do so, especially if a method suited to ordinal data is available.
Hello, In newer versions of SPSS, you have to specify that your variable is ordinal in the Variable View. That's likely the issue. If the variable is specified as "scale" or "nominal," then the non-parametric drop-down menus won't allow you to enter the variable into an ordinal routine.
Thanks for the explanation. One question: if i want to compare a nominal variable between groups, but the nominal variable has more than 2 options (eg blood type) would I need to use Mann Whitney as in your ordinal variable example?? Or what do you suggest? Thanks
Hello - no - if your outcome variable is multiple "unordered" categories-as it sounds in your example-and your independent variable is multiple "unordered" categories, you could just use Chi square. However, if your outcome variable was an ordered categorical variable (i.e., an ordinal level of measurement), and your independent variable was a two-level unordered categorical variable, then you could use the Mann-Whitney test.
Thanks for the lecture! My question is what should I do if I want to know the group difference (control and treatment group) between pre-and post-test. Or is it reasonable to do so because it's not normally distributed? Thanks!
Your data do not need to meet an assumption of normality for non parametric tests. For related samples use the Wilcoxon test with ordinal data or the McNemar test with nominal data. It is not possible to analyze a 2 x 2 design with these tests. One option is to test if the groups are comparable on the pre-test, and if they are, then focus on the differences at post-test.
Hi, thank you for the video I have a doubt I would like to ask which test in spss is suitable for evaluate the level of public awareness and knowledge about the disposal of medication and its effect on Environmental and Public Health: (A Survey Among UAE ) The question to apply on SPSS are : What is your gender: Female Male What is your educational level? Illiterate. School. Bachelor. Master. Doctorate. What is your Occupation level? Employee. Student. Housewife. Retiree. Unemployed. How do you dispose of expired drugs? In the garbage. Flushing. Returned to Pharmacy. Storing. Always used till finished. Can I use the spearman correlation test Or Chi square test Thanks i would appreciate your response
Literally been trying to find out how to get the p-value between 2 groups of nominal data for aaages!! thank you heaps. very helpful
Yes, this is exactly what I needed! Thanks! 💯
excellent thank you for comparing multiple tests for one data set!
Thank You for the lecture. My question would be which test should I use if I want to see the p-value of the change comparing the same nominal data before intervention and after intervention (nominal variable has 4 options but respondent can choose just one of them both times - before and after intervention)?
Hello, I believe the Cochran's Q test would be suitable for your example.
@@petemiksza Thank You. I tried it, yet instead of p-value system shows a letter ,,a", which is described as ,,all test fields do not have exactly two values" below the table of hypothesis.
@@VaivaGegužinskaitė oh, I see, I’m sorry the Cochran Q test works for comparisons of a dichotomous outcome on more than two occasions - you may want to try the “marginal homogeneity test” : www.statisticssolutions.com/free-resources/directory-of-statistical-analyses/mcnemar-marginal-homogeneity-sign-wilcoxon-tests/
Thank You once again. Tried it as well. But I guess that this test fits better for ordinal data instead of nominal.
Thanks a lot! In my SPSS, when I double click on the graph to get more details about the statistic, nothing new appeared :( can anyone help?
What version of SPSS are you using?
@@petemiksza Version 27
@@anacatarinasantos7752 In this video, I used an older version, in which you had to "double-click" on the summary output for some non-parametric tests to get more detailed information. In versions 27 and 28 the more detailed information that only appeared in previous versions when you "double clicked" the output are now presented by default - all the information should be there. In other words, there's nothing else to see.
Hi, may I know do u have any suggestions for me to do correlation?
I had actually collected the agreement level on the benefits with 5 point likert scale. On the other hand, i do have data on their willingness to use with yes or no choice. May I know which method is suitable?
Hello - if you have one dichotomous variable representing independent groups responses - yes or no - and one ordinal variable (e.g., a rating scale item), then you can use a Mann-Whitney U test to compare the rating scale responses as a function of whether participants responded yes or no to the dichotomous item.
If you can treat the ordinal categories of the rating scale item as nominal, then you could use the “Cramer’s V” coefficient.
There’s also a specific coefficient that I do not know much about that is suited for these data called Freeman’s theta.
Thank you for this video!! ♥️♥️ I just want to confirm if I should use Chi-square in correlating nominal data with four categories to ordinal data with three categories? Some said that I should use point biserial correlation with this one.
Yes, you could use Chi square in that circumstance. In doing so, you would be treating your ordinal variable as a purely nominal variable. I do not believe a point-biserial correlation would be appropriate in this case. It is typically used when one variable is dichotomous (i.e., two categories) and the other variable is scalar (i.e., interval- or ratio-level).
Hello
Thanks for the informative lecture. One question . I need your help and clarification in something. I am conducting a study to determine the difference in physical activity pattern before and after covid 19. If I want to compare before and after for multiple outcome variables ( never
Exercised,once, twice, and three times) which statistical test is best to compare the difference before and after?
Hello, It sounds like you want to compare responses the same group of participants made to an item asking how often they exercised with response options of never, once, twice, or three times on multiple occasions (i.e., pre- and post-COVID-19). If you treat those data as "interval/ratio-level" measurements, then you could use a dependent samples t-test to look at whether the mean difference between the two time points is significant. If you treat those data as "ordinal-level" measurements, then you could use a Wilcoxon test to look at whether the difference between the responses at the two time points is significant. Given there are only 4 possible outcomes, you might consider treating those data as "ordinal," especially if you have a small N; even though-technically-the response options represent "equal intervals" (i.e., 0-1-2-3), the responses seem likely to be distributed in a fashion that is more similar to an ordinal measurement.
Thank you for your quick response! I appreciate it.
Yes, I am comparing the resposes for the same group but before and after a certain period of time.
I have more than 4 possible outcomes. The outcomes I have are:
Never
Once
Twice
Three time
More than three times
I have a large sample of 370 participatns. I wonder if this kind of data can be treated as ordinal data? Can it be?
If yes, then I can go for wilxon sign rank test?
Another related question. What if I also have other multiple categorical variables and I also want to test the difference before and after, such as the following:
How many minutes did you exercise in a typical week before covid 19 pandemic?
Less than 30 minutes
30 minutes
30 minutes - less than 1 hour
1 hour - less than 2 hours
2 hours
And then the same question again "How many minutes did you exercise last week? "
Which test can also be suitable for this type of data?
Thank you very much
@@halahala3524 Hello, each of the response modes you described in your reply include "ordinal" categories (i.e., more than three times, less than 30 minutes) - so, I would definitely recommend using a non-parametric test for comparing changes in responses over time with those items - for two measurement occasions, the Wilcoxon test is appropriate.
Thanks for the clarification
Is likert scale questionnaire data always ordinal and be analyzed via
non-parametric tests?
Some argue that if a Likert scale variables involves enough response options (e.g., more than 5 scale points) and the distribution of the values roughly resembles a normal distribution, then you can apply parametric tests. This is generally the rationale many psychologists use. However, many statisticians would say it is never appropriate to do so, especially if a method suited to ordinal data is available.
Hi why cant I put ordinal data in test field
Hello, In newer versions of SPSS, you have to specify that your variable is ordinal in the Variable View. That's likely the issue. If the variable is specified as "scale" or "nominal," then the non-parametric drop-down menus won't allow you to enter the variable into an ordinal routine.
Thanks for the explanation. One question: if i want to compare a nominal variable between groups, but the nominal variable has more than 2 options (eg blood type) would I need to use Mann Whitney as in your ordinal variable example?? Or what do you suggest? Thanks
Hello - no - if your outcome variable is multiple "unordered" categories-as it sounds in your example-and your independent variable is multiple "unordered" categories, you could just use Chi square. However, if your outcome variable was an ordered categorical variable (i.e., an ordinal level of measurement), and your independent variable was a two-level unordered categorical variable, then you could use the Mann-Whitney test.
Thanks for the lecture! My question is what should I do if I want to know the group difference (control and treatment group) between pre-and post-test. Or is it reasonable to do so because it's not normally distributed? Thanks!
Your data do not need to meet an assumption of normality for non parametric tests. For related samples use the Wilcoxon test with ordinal data or the McNemar test with nominal data. It is not possible to analyze a 2 x 2 design with these tests. One option is to test if the groups are comparable on the pre-test, and if they are, then focus on the differences at post-test.
@@petemiksza Thank you very much for your timely reply! This is super helpful. Happy Holidays!
Hi, thank you for the video
I have a doubt
I would like to ask which test in spss is suitable for evaluate the level of public awareness and knowledge about the disposal of medication and its effect on Environmental and Public Health: (A Survey Among UAE )
The question to apply on SPSS are :
What is your gender:
Female
Male
What is your educational level?
Illiterate.
School.
Bachelor.
Master.
Doctorate.
What is your Occupation level?
Employee.
Student.
Housewife.
Retiree.
Unemployed.
How do you dispose of expired drugs?
In the garbage.
Flushing.
Returned to Pharmacy.
Storing.
Always used till finished.
Can I use the spearman correlation test
Or
Chi square test
Thanks i would appreciate your response
Thank you!! very helpful for my thesis lol