Compute t-test for a root mean square error

I would like to compute a t-test for two vectors. More specifically, test whether

RMSE_1-RMSE_2 is equal to zero against the alternative that the difference is negative. I have a vector of RMSE_1 and RMSE_2 that were computed for different sample sizes. The sample size was chosen as an decreasing fraction(1/2:1/20) of the total sample size.

As a result I have two 19*1 vector and would like to get the confidence interval and test result for this statistic.

Any help or function how can I do that? the ttest2 function tests whether the two vectors have equal means or not which is not what I want to implement.

Answer 1

[Star Strider]

‘ttest2 function tests whether the two vectors have equal means or not which is not what I want to implement’

If I understand what you want to do as you described it, ttest2 is what you want to implement if you want to use your two vectors in an unpaired design.

If you want to subtract them first, and test if that mean is different from zero, use the ttest function for a one-sample test, or both original vectors a paired design.

You must decide the appropriate test for your data.