topic Univariate Hodges-Lehmann Estimator? in Statistical Procedures

Univariate Hodges-Lehmann Estimator?

xtc283x — Wed, 11 Jul 2018 10:24:48 GMT

Proc NPAR1WAY in v9.4 contains an option for the two sample Hodges-Lehmann location estimate but, based on my survey of the SAS Support literature, there does not appear to be a procedure supporting a univariate HL estimate nor does there appear to be a computationally feasible approach to producing on with large n.

Is this correct?

Re: Univariate Hodges-Lehmann Estimator?

Rick_SAS — Thu, 12 Jul 2018 13:05:58 GMT

How big is your sample size, N?

Re: Univariate Hodges-Lehmann Estimator?

xtc283x — Thu, 12 Jul 2018 13:09:42 GMT

Quite large...millions of records. Thank you!

Re: Univariate Hodges-Lehmann Estimator?

Rick_SAS — Thu, 12 Jul 2018 13:29:58 GMT

I'm sure you are aware that the H-L estimator requires N(N+1)/2 comparisons to estimate the location parameter in the population. In contrast, the sample median estimates the location parameter much faster and is also a robust estimator. For a data set with millions of records, I would expect the median and the H-L estimate (the pseudo-median) to be very close.