Whats the suitable test for Comparison?

[Olalere Oluwaseyi]

Hello Everyone,

I will like to know the kind of test suitable for comparison of set of groups.

Precisely, I am working on commercial banks and I'm trying to group my data into Big Banks and non-big banks according to their assets.

So, the question is, what type of test is suitable for this kind of scenario.

Many thanks.

[Clyde Schechter]

I find your question too broad, and unclear. It is unclear in that I cannot tell whether you are asking for a way to actually group the data into big and non-big banks, or for a way to compare them on other attributes once you have done so. In either case, there are many, many possible answers to both of those questions and they depend on your research goals and on the nature of the data you have to work with. I suggest you read the Forum FAQ for excellent advice on how to pose questions that maximize your chances of getting a timely and helpful response, and then repost with a more specific, focused question.

[Olalere Oluwaseyi]

Thank you Clyde.

Precisely, my question is not on how I can group my data but on how I can comparison after grouping.

I understand some researchers use Two sample t-test for comparison, I just wanna know which other test is suitable.

Many thanks.

[Clyde Schechter]

As I said before, there are many such tests. And which might be appropriate depends on the nature and distributions of the attribute(s) that you want to compare across groups. Without information about that, attempting to answer this question would be writing the better part of a statistics text book, not a post in an online forum.

[Joseph Coveney]

I understand some researchers use Two sample t-test for comparison, I just wanna know which other test is suitable.

Here's one:

sysuse auto
dotplot price, over(foreign) median center bar mcolor(black) msize(vsmall) ylabel( , angle(horizontal) nogrid)

[Weiwen Ng]

Thank you Clyde.

Precisely, my question is not on how I can group my data but on how I can comparison after grouping.

I understand some researchers use Two sample t-test for comparison, I just wanna know which other test is suitable.

Many thanks.

Two sample t-test is likely to be it for continuous outcomes (or outcomes that are proportions). I suspect that the variances of your groups could be unequal, and if so, you would probably be better invoking the appropriate option. So, using fictitious variable names for your data:

ttest y, by(tbtf) unequal

Where tbtf is an indicator for banks that are, as they say, too big to fail. This should raise the question: if a bank is too big to be allowed to fail, is it also too big to be allowed to exist? But that's a side question!

If outcomes are skewed, some people recommend the Krukskal Wallis test, which is (I believe) essentially a test that the medians of two groups are equal. Or the signed rank test, albeit I believe this is limited to matched data (unlikely to occur in your case unless you somehow matched a TBTF bank with a smaller bank). The code that Joseph showed isn't a statistical test, but it does plot the medians (+++s) of the two groups, plus the 25th and 75th percentiles (the different colored ---s).

If outcomes are categorical (including binary, ordered categorical, and unordered categorical), then the chi square test is the logical one to use. If there is a very small frequency of observations in one or two cells (I've heard fewer than 5 cases in one cell), then you can probably use Fisher's exact test instead.

If you want to go into more detail, then as Clyde mentioned, that could take up the better part of a full text book. If you want to post comparisons for table 1 of a paper, then you honestly only need to know the t-test plus the chi-square test, as you will be adjusting for all the differing characteristics in some sort of regression. We tend to only need rough descriptive statistics for table 1.

[Dan Palmer]

The UCLA Institute for Digital Research and Education has a nice matrix that lays out recommended tests based on characteristics of the data.

https://stats.idre.ucla.edu/other/mult-pkg/whatstat/

[Olalere Oluwaseyi]

Thank you for the kind responses.
They've been very helpful.