Chi square with ties

[PLR1]

Hi
So I am conducting a kwallis test which produces two chi square outputs, one with ties.
My question is, I am getting two completely different p values, so not sure what this means....any suggestions?

kwallis v3ssiorgan, by(state)

Kruskal-Wallis equality-of-populations rank test

+------------------------+
| state | Obs | Rank Sum |
|-------+-----+----------|
| A | 5 | 181.50 |
| B | 17 | 568.50 |
| C | 1 | 52.50 |
| D | 19 | 673.50 |
| E | 6 | 315.00 |
|-------+-----+----------|
| F | 1 | 12.00 |
| G| 19 | 876.00 |
| H | 13 | 642.00 |
+------------------------+

chi-squared = 8.751 with 7 d.f.
probability = 0.2711

chi-squared with ties = 14.344 with 7 d.f.
probability = 0.0454

many thanks
Phil

[Nick Cox]

So, the result of your analysis depends entirely on the treatment for ties. Not a good state to be in. Proceed with caution. And please do change your identifier from "PLR1". This is explained in the FAQ Advice Section 6.

[Melissa Wilson]

A similar situation continually arises with my dataset where, it may be significant both times, but borderline without ties and highly significant with ties. i feel uncomfortable with this and am not sure what "proceed with caution" means other than what I already feel.

[Nick Cox]

Try this then: if your result is totally sensitive to how tries are treated, you don't have enough information to decide what to think. Try a different method, e.g. to establish a significance level by simulation, or wonder whether deciding whether a result is significant at conventional levels is really the scientific goal.