Skip to: Main Content / Navigation

  • Facebook
  • Twitter
  • LinkedIn
  • Add This

All Forums > Statistical Analysis Questions

Univariate Analysis

When it's time to test significant on top 2 box or mean. I am quite confuse when to you Z, Chi-square, and F (Anova).

Thank you in advance!!

Compared to what?

To effectively answer your question, you must be more specific with regard to what question you are trying to answer. For example:

- Are you comparing the significance of a top 2 box proportion as compared to 0, to a previous study, the top 2 box for males versus females, or what? It makes a differences, although generally you will want to be looking at the binomial or a chi-square test for this sort of issue.

- Same questions apply for means, although here F-tests and t-tests will be the "usual suspects" that you will look at.

Univariate Analysis

Hi Mangen,

Thank you so much for your clarification.

May I confirm my understanding as follow?

if compare top 2 for the independent samples as nominal data, say male v.s. female it must be chi square

If compare top 2 for the same sample, say N=100 test 2 products and rating on both products, it must be proportion t-test.

If compare mean for 2 independent samples, say female v.s. male, it must be......????.

If compare mean for 2 dependent samples, say N=100 test 2 product and rating on both product, it must be compare mean t-test, if 3 product it should be f-test.

Thank you so much in advance

Concerned

Without knowing exactly what you are doing, I am concerned about giving too much in the way of detailed advice. Why? Because there are further questions that are raised with this new post, and my experience suggests that tends to serve as an indicator that there are other things that are "going on" that have not been communicated which could change my response -- perhaps dramatically! Furthermore, the wording of your post fairly clearly suggests that you are looking for simple rules, and in some cases the rules are not that simple. Nonetheless, I will try to answer your questions:

"...if compare top 2 for the independent samples as nominal data, say male v.s. female it must be chi square." Yes chi square will work, as would an independent sample t-test comparing proportions.

"...If compare top 2 for the same sample, say N=100 test 2 products and rating on both products, it must be proportion t-test." Assuming that you intend to compare the proportion top 2 box for Product A to the proportion top two box for Product B in a single sample where all respondents have rated both products, then the simplest test is to take a difference of the proportions and test whether that difference score is significantly different from 0 using a t-test. However, you might want to reformulate this problem as a multivariate analysis of variance (MANOVA) which will allow you to test the three or more product case, as well as the two product case.

"...If compare mean for 2 independent samples, say female v.s. male, it must be......????." Either a simple t-test or an unbalanced ANOVA f-test would be applicable here.

"...If compare mean for 2 dependent samples, say N=100 test 2 product and rating on both product, it must be compare mean t-test, if 3 product it should be f-test." Again assuming that you intend to compare the average for Product A to the average for Product B in a single sample where all respondents have rated both products, then the simplest test is to take a difference of the proportions and test whether that difference score is significantly different from 0 using a t-test. However, you might want to reformulate this problem as a multivariate analysis of variance (MANOVA) which will allow you to test the three or more product case, as well as the two product case.

T-test versus non-parametric techniques

I have matched pre and posttest data. Much of the data consist of ratings on a four point scale (Strongly agree, etc). Most respondents are at one end of the scale (ie 87% either strongly agree or agree). Since this is a multi-year study, with many questions, I would like to report means. Is it kosher to use a ttest to determine if there are significant differences in the means - pre and post? (approximately 90 matched subjects)

Another portion of the survey computes a score for each respondent, based on weighted responses to a series of questions. Must this data be normally distributed to do a ttest?
And if not, which test to use? (66 subjects matched for these scores)

I read about non-parametric tests, but other than chi-square, I don't know too many "regular" researchers that use them.

Thanks for any input.

Reference Article

While it might not answer all your questions, there is an article in the May 2002 Review, Data Use Section, called "Nonparametric Tests: Sturdy Alternatives" that might be of interest. It also is available for download at my Web site - See the Site Map.