e() results

[paulvonhippel]

I'm running the loneway command and looking at the e() results. Some things surprise me:
(1) I run loneway once and look at the e() results. Then I run loneway again on a different subset of data. The e() results haven't changed; they are the results from my first loneway run, not my second run.
(2) Concerned, I run "eresults clear" and then run loneway again. Now the e() results are empty, which surprises me because I expected them to reflect my last loneway run.

What's going on? Is this normal behavior? I haven't used e() results before, so I don't know if I'm seeing a bug or just not understanding how e() results are supposed to work.

Many thanks.

Best,
Paul

[Richard Williams]

Loneway doesn't ereturn anything. See -help loneway-. You probably saw the e() results left over from some earlier estimation command.

[paulvonhippel]

No, the e() results that I saw were from loneway. They matched what I saw in the Results window.

[paulvonhippel]

Oh, no, wait, you're right. The e() results that I saw were leftover from an earlier anova command.
Mystery solved -- thanks.

[Richard Williams]

You sure they weren't r() results? The first two lines of loneway are

*! version 3.1.6  15may2009
program define loneway, rclass byable(recall) sort

It is an R class command, not E class. And the help for it says

Stored results

    loneway stores the following in r():

    Scalars   
      r(N)           number of observations
      r(rho)         intraclass correlation
      r(lb)          lower bound of 95% CI for rho
      r(ub)          upper bound of 95% CI for rho
      r(rho_t)       estimated reliability
      r(se)          asymp. SE of intraclass correlation
      r(sd_w)        estimated SD within group
      r(sd_b)        estimated SD of group effect

[paulvonhippel]

A followup question, as long as we're here...
loneway reports the reliability of an average, "evaluated at n=5.38" for example. How is that value, 5.38, calculated?
I have to recreate it because it's not in the r() results!

[Richard Williams]

On p. 1099. there is a formula for something called g, and on p. 1101 it says loneway uses t = g. I suspect that is it but I haven't calculated it. If that isn't it, then the answer may be buried elsewhere in the formula and methods section.

[Richard Williams]

Ok, I think I was right. See the pages I gave before or this won't make any sense:

. webuse nhanes2f, clear

. loneway height race

              One-way Analysis of Variance for height: height (cm)

                                              Number of obs =     10337
                                                  R-squared =    0.0071

    Source                SS         df      MS            F     Prob > F
-------------------------------------------------------------------------
Between race           6883.2221      2     3441.611     37.14     0.0000
Within race            957629.15  10334     92.66781
-------------------------------------------------------------------------
Total                  964512.37  10336    93.315826

         Intraclass       Asy.        
         correlation      S.E.       [95% Conf. Interval]
         ------------------------------------------------
            0.03054     0.03772       0.00000     0.10448

         Estimated SD of race effect             1.708707
         Estimated SD within race                9.626412
         Est. reliability of a race mean          0.97307
              (evaluated at n=1147.02)

. tab1 race

-> tabulation of race  

   1=white, |
   2=black, |
    3=other |      Freq.     Percent        Cum.
------------+-----------------------------------
      White |      9,051       87.56       87.56
      Black |      1,086       10.51       98.07
      Other |        200        1.93      100.00
------------+-----------------------------------
      Total |     10,337      100.00

. di (10337 - (9051^2 + 1086^2 + 200^2)/10337)/2
1147.0239

.

The display command at the end gives the formula for "evaluated at n="

[Richard Williams]

Here is some all purpose code (assuming no weighting). If you had some mad desire to constantly double-check Stata, you could probably make this more generalizable.

capture program drop cloneway
program cloneway
    args y x
    loneway `y' `x'
    quietly tab1 `x' if !missing(`y'), matcell(freq)
    mat cloneway = (r(N)  - ( freq' * freq)/r(N))/(r(r) - 1)
    mat list cloneway
end

* Examples
webuse auto7, clear
cloneway mpg manufacturer_grp

webuse nhanes2f, clear
cloneway height health

Here is the output:

. capture program drop cloneway

. program cloneway
  1.         args y x
  2.         loneway `y' `x'
  3.         quietly tab1 `x' if !missing(`y'), matcell(freq)
  4.         mat cloneway = (r(N)  - ( freq' * freq)/r(N))/(r(r) - 1)
  5.         mat list cloneway
  6. end

.
. * Examples
. webuse auto7, clear
(1978 Automobile Data)

. cloneway mpg manufacturer_grp

              One-way Analysis of Variance for mpg: Mileage (mpg)

                                              Number of obs =        74
                                                  R-squared =    0.5507

    Source                SS         df      MS            F     Prob > F
-------------------------------------------------------------------------
Between manufactur~p    1345.588     22    61.163092      2.84     0.0011
Within manufactur~p    1097.8714     51    21.526891
-------------------------------------------------------------------------
Total                  2443.4595     73    33.472047

         Intraclass       Asy.        
         correlation      S.E.       [95% Conf. Interval]
         ------------------------------------------------
            0.36827     0.13679       0.10017     0.63636

         Estimated SD of manufactur~p effect     3.542478
         Estimated SD within manufactur~p        4.639708
         Est. reliability of a manufactur~p mean  0.64804
              (evaluated at n=3.16)

symmetric cloneway[1,1]
           c1
r1  3.1584767

.
. webuse nhanes2f, clear

. cloneway height health

              One-way Analysis of Variance for height: height (cm)

                                              Number of obs =     10335
                                                  R-squared =    0.0245

    Source                SS         df      MS            F     Prob > F
-------------------------------------------------------------------------
Between health         23584.908      4     5896.227     64.75     0.0000
Within health          940695.15  10330     91.06439
-------------------------------------------------------------------------
Total                  964280.06  10334    93.311405

         Intraclass       Asy.        
         correlation      S.E.       [95% Conf. Interval]
         ------------------------------------------------
            0.03101     0.02301       0.00000     0.07611

         Estimated SD of health effect           1.707074
         Estimated SD within health              9.542766
         Est. reliability of a health mean        0.98456
              (evaluated at n=1992.09)

symmetric cloneway[1,1]
           c1
r1  1992.0926