# Re: st: RE: Nonparametric test

 From David Airey To statalist@hsphsun2.harvard.edu Subject Re: st: RE: Nonparametric test Date Tue, 7 Oct 2008 11:24:16 -0500

I sincerely doubt Nigussie would be posing these questions if Nigussie were in charge of a mutlinational survey of income levels, and so pointing out the upper bar for statistical correctness is didactic, but about as useless as suggesting a chi-square statistic is adequate.
```
-Dave

On Oct 7, 2008, at 10:37 AM, Steven Samuels wrote:

```
```
```
Unless you took care to identify all the person you could have sampled, and you ensured each of those persons had an equal likelihood of being sampled, it doesn't sound to me like you are in a position to use survey statistics.
```
```
```
```
Nigussie may have have survey data even if neither of these conditions are true:
```
```
1) Rarely do surveys have a list of individuals who could have been sampled. Simple case: a three stage area sample: 1. blocks; 2. households, 3. individuals. The first list is a list of blocks. After units are drawn at one stage, only units eligible for drawing at the next stage are listed.
```
```
2) Most multi-stage samples do not maintain equal probabilities of selection for the last stage units. It can be done, however. For example, an area sample can make the probability of selecting each household the same. If information is gathered from or about all individuals in a household, each will inherit the household's probability.
```
-Steve

On Oct 7, 2008, at 10:22 AM, David Airey wrote:

```
But maybe here is a case where the person asking the question is not familiar with statistics generally? Any answer might be helpful. Advice given here is just advice, and there is no promise it is correct.
```
```
If you have a table of two categorical variables (e.g., country and income group) and you can make a table of frequencies of each country and income group combination, a chi-square statistic is one way to ask if there is an association between the two variables. It is a nonparametric test.
```
```
A confusing aspect of your presentation is the variable individual, that it is numbered in a way that suggests the same individual is in multiple countries, and that your are talking about paired analyses. That may throw off people into thinking your design is more complicated than it really is. My guess is you have a simple listing of different people from different countries with associated incomes. Unless you took care to identify all the person you could have sampled, and you ensured each of those persons had an equal likelihood of being sampled, it doesn't sound to me like you are in a position to use survey statistics.
```
See the command -tabulate twoway- for a simple solution.

See the command -survey- if I'm wrong.

-Dave

On Oct 7, 2008, at 7:22 AM, Nick Cox wrote:

```
```Your data structure is now clear. Thanks for that.

```
I remain totally in the dark on what kind of test and what kind of hypothesis you have in mind. You ask for a paried test but even a guess that that is a typo for paired does not help me as I do not see here a paired comparison of any kind.
```
```
What's more, advice is difficult if not foolish in the absence of information on survey design.
```
```
Sorry not to be able to add more, but you didn't answer several questions in my earlier post.
```

Nick
n.j.cox@durham.ac.uk

nigussie Tefera

```
I am very sorry for stating my question in an ambiguous form. Here, is an example
```
country    individual    income group
x                1            2
x                2            1
x                3            2
x                4            3
.                  .            .
.                  .            .
.                  .            .
y                1            1
y                2            3
y                3            1
y                4             2
.                .             .
.                .             .
.                .             .
y                1            3
y                2            2
y                3            1
y                4             1
.                .             .
.                .             .
.                .             .
```
Now I want to generate the following table with paried test (compared all countries)
```
Income group                country
```
X y z w paried test 1 count count count count ? 2 count count count count ? 3 count count count count ?
```

```
If I want to fill all ? (question marks) what stata command should I use? Note: I have no problem in producing count. By the way, count can be either row or column percentages. I hope this will clarify the issues
```
Nick Cox <n.j.cox@durham.ac.uk>

```
I don't get a clear picture of what your data look like. We have countries, groups and individuals floating around in perfect fuzziness. Let me try again.
```
1. Do your data look like this? i.e. one observation, one country.

Country  Income group
x            1
y            2
z            2
w            3

```
2. Or do you have replicates (e.g. areas, individuals) within each country?
```i.e. several observations for each country.

3. Or do you have something else? If so, give an example.

```
Also, what do you mean when you say that you want a separate test for each income group?
```
No difference in what?

Nick
n.j.cox@durham.ac.uk

nigussie Tefera

Many thanks!

```
My data looks like eactly what you stated but country x has high, middle and low income groups (individuals) and the same is applied for country y, z and w. Now, I want to test, whether there is statistically significant difference among high income groups across countries. The same is also true for mindle and low income groups. In short: the null hypothesis is no difference among countries for low, middle and high income groups (I need separate test for each income groups)
```
Nick Cox <n.j.cox@durham.ac.uk>

Do your data look like this?

Country  Income group
x            1
y            2
z            2
w            3

```
Or do you have replicates (e.g. areas, individuals) within each country?
```
And what hypothesis are you testing?

nigussie Tefera

```
I have two categorical variables: namely incomecat (1=high, 2=middle and 3=low income) and country ( labeled as (x, y, z, w)). Suppose that each country has high, middle and low income group. If I want to run paired test for each income group across countries, what stata command should I use?
```
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```
```
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```
```
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```
```
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```