When I try to run a Mann Whitney Tes, I get the following error

Warnings

Frequency weight data values must be positive integers.

Positive non-integer frequency weight data values are encountered. These values are rounded to the nearest integers for the analysis.

How can I resolve this? If I multiply the weight to a number to remove fractions, the significance is exaggerated. is this the right thing to do.

The nonparametric procedures need integer weights for the computation. If your weights arise from a nonrandom sample, you might consider the statistics available in the complex sample procedures and/or use parametric tests.

You might also see whether the test conclusions vary importantly when you multiply the weights by 10.

Many thanks Jon. When I tried to multiply the weights by 10, the non-parametric test conclusions varied greatly when compared to parametric tests (with original weights). Almost all variables tested were significantly different in non-parametric tests. I think I will use parametric tests instead.

I can't offer a definitive answer without knowing more about the situation, but you might want to consider using effective base weighting and a column proportions or column means test in CTABLES.

Hi Jon, How do you use effective base weighting in SPSS. I have read about it but I do not know how I can create it in SPSS and apply it to carry out analysis as we do with weighting. Are the links that could be useful. All I want is to run a non-parametric test with weights. I am not intending to use CTABLES.

Effective base weighting is only available in CTABLES, which supports column proportion tests and column means tests. Nothing nonparametric.

Thanks Jon. that's good to know. Can I run nonparametric tests in complex samples?

The Complex Sample procedures do not have an equivalent of the nonparametric procedures, but maybe you can formulate the question in a way that their descriptives or GLM procedures can answer.

Look at the descriptions in the Command Syntax Reference for details.

The procedure names all start with CS, so you will see them together in the CSR.

You can get nonparametric tests from R if you have details of the sampling design, but it would be a bit of work. You can read about that here

Survey Data Analysis with R (ucla.edu)

It has a section on non-parametric [sic] tests, but you have to set up the sample design specifications first.