SE=0 shown as really small values `bydomain`

[smishr]

I noticed negligible standard errors (almost 0) in many domain estimation cases. like here

julia> mean(:api00, :cname, bclus1)
11×3 DataFrame
 Row │ cname        mean     SE
     │ String15     Float64  Float64
─────┼────────────────────────────────────
   1 │ Santa Clara  732.077  58.2169
   2 │ San Diego    659.436   2.66703
   3 │ Merced       519.25    2.28936e-15
   4 │ Los Angeles  647.267  47.6233
   5 │ Orange       710.563   2.19826e-13
   6 │ Fresno       472.0     1.13687e-13
   7 │ Plumas       709.556   1.26058e-13
   8 │ Alameda      669.0     1.27527e-13
   9 │ San Joaquin  551.189   2.1791e-13
  10 │ Kern         452.5     0.0
  11 │ Mendocino    623.25    1.09545e-13

Are these just floating point 0's? should they be printed as "0.0" like for Kern ? But im pretty sure these are not lonely PSU cases, there is more than 2 values to calc a variance?

[smishr]

@ayushpatnaikgit

[sayantikaSSG]

This is not a floating-point error, but actual standard error of mean calculation. should we just approx the error to 0?

[smishr]

This is not a floating-point error, but actual standard error of mean calculation. should we just approx the error to 0?

this is negligible error. i think those domains have single value. calculation of variance from a single number has options- Nan, missing, or just 0.

[smishr]

@sayantikaSSG @itsdebartha can you check that the really small SE values are coming from domain with size=1. If so, then add condition in bydomain to give NaN/NA value for these domains. When printing, these domains should also print NA.

@ayushpatnaikgit @codetalker7 Im assuming it is fine to have NA values in numeric vectors in Julia? There are no performance penalties or computation difficulties?