# Thread: Parametric test results vs Non-parametric test results - interpretation help please?

1. ## Parametric test results vs Non-parametric test results - interpretation help please?

Hello, thank you so much for your help.

I am currently working on an article about trends in streamflow in watersheds with different land use change over time. For this I need to analyze trends in streamflow parameters over time.

Both non-parametric and parametric tests were applied to the same sets of data (time series).

Used:
Non-parametric tests : Mann-Kendall, Theil-Sen Estimator;
Parametric test: simple linear regression with 1 independent variable and an F-test to test for significance.

I have no original data (the guy who had it disappeared ). Due to the nature of the subject it is hard to say whether or not the trend is likely to be linear and monotone , and what the distribution may be.

The problem is as follows: the parametric tests and the non-parametric test show opposite results for the same data sets in some cases (meaning one test finds a significant trend in data and the other one does not). I an really confused as to how to interpret this - which test should I trust. Any suggestions would help.

I have not done any kind of statistics in 3 years so I am really rusty. I will appreciate any help.

Sincerely,

Olia

P.S. Not sure this is the right section to post this. If you think of a better place please tell me.

2. ## Re: Parametric test results vs Non-parametric test results - interpretation help plea

I dont know if you can really answer this without the data. Each test is used in different situations. If your data is very non-normal, or the sample is small, you might want to use Mann-Kendall. Is there anyway you can find out anything more about the data at all? Or even find some similiar data and try to work out what kind of distributions these thigns usually follow?

Without the original data its hard to check if your data violates any of the assumptions of the tests. For example, for linear regression the errors need to be normally distributed.