log type:  text

 opened on:  14 Nov 2005, 11:02:19

 

. use "C:\AAA Miker Files\newer web pages\soc_388_notes\clogg and eliason data.dta", clear

 

. drop _x_*

 

. table color  labor, contents (sum uwcount sum wncount mean weight) by (sex)

 

----------------------------------------------

sex and   |               labor              

color     | unemployed   part-time       other

----------+-----------------------------------

male      |

    white |       3511        4227       31467

          |       3530        4183       31131

          |       1431        1408        1408

          |

    black |        604         356        2245

          |        815         462        2783

          |       1921        1849        1764

          |

    other |        165         157         924

          |        119         124         797

          |       1029        1124        1228

----------+-----------------------------------

female    |

    white |       2281        7833       18945

          |       2234        7559       18704

          |       1394        1373        1405

          |

    black |        545         563        2132

          |        653         644        2498

          |       1705        1627        1668

          |

    other |         89         216         725

          |         64         162         574

          |       1029        1070        1127

----------------------------------------------

 

. *This is the data table from Clogg and Eliason, with unweighted counts, weighted normalized counts, and mean cell weights.

. set linesize 79

 

. desmat: poisson  uwcount  labor*sex=dev(3)*dev(2) labor*color=dev(3)*dev(3) sex*color=dev(2)*dev(3)

-------------------------------------------------------------------------------

   Poisson regression

-------------------------------------------------------------------------------

   Dependent variable                                                  uwcount

   Optimization:                                                            ml

   Number of observations:                                                  18

   Initial log likelihood:                                          -81627.074

   Log likelihood:                                                    -123.390

   LR chi square:                                                   163007.367

   Model degrees of freedom:                                                13

   Pseudo R-squared:                                                     0.998

   Prob:                                                                 0.000

-------------------------------------------------------------------------------

nr Effect                                                    Coeff        s.e.

-------------------------------------------------------------------------------

   uwcount

     labor

1      unemployed                                           -0.677**     0.018

2      part-time                                            -0.433**     0.017

     sex

3      male                                                 -0.018*      0.009

     labor.sex

4      unemployed.male                                       0.161**     0.009

5      part-time.male                                       -0.347**     0.007

     color

6      white                                                 1.852**     0.011

7      black                                                -0.364**     0.014

     labor.color

8      unemployed.white                                     -0.282**     0.018

9      unemployed.black                                      0.340**     0.022

10     part-time.white                                       0.193**     0.017

11     part-time.black                                      -0.229**     0.022

     sex.color

12     male.white                                            0.080**     0.009

13     male.black                                           -0.097**     0.011

14   _cons                                                   7.053**     0.011

-------------------------------------------------------------------------------

*  p < .05

** p < .01

 

. poisgof

 

         Goodness-of-fit chi2  =  86.53056

         Prob > chi2(4)        =    0.0000

 

. poisgof,pearson

 

         Goodness-of-fit chi2  =  89.79915

         Prob > chi2(4)        =    0.0000

 

. *This is Clogg and Eliason's model 1

. desmat: poisson   wrongcount labor*sex=dev(3)*dev(2) labor*color=dev(3)*dev(3) sex*color=dev(2)*dev(3)

-------------------------------------------------------------------------------

   Poisson regression

-------------------------------------------------------------------------------

   Dependent variable                                               wrongcount

   Optimization:                                                            ml

   Number of observations:                                                  18

   Initial log likelihood:                                          -1.137e+08

   Log likelihood:                                                  -71806.645

   LR chi square:                                                    2.273e+08

   Model degrees of freedom:                                                13

   Pseudo R-squared:                                                     0.999

   Prob:                                                                 0.000

-------------------------------------------------------------------------------

nr Effect                                                    Coeff        s.e.

-------------------------------------------------------------------------------

   wrongcount

     labor

1      unemployed                                           -0.689**     0.001

2      part-time                                            -0.435**     0.000

     sex

3      male                                                  0.013**     0.000

     labor.sex

4      unemployed.male                                       0.165**     0.000

5      part-time.male                                       -0.340**     0.000

     color

6      white                                                 1.858**     0.000

7      black                                                -0.133**     0.000

     labor.color

8      unemployed.white                                     -0.263**     0.001

9      unemployed.black                                      0.383**     0.001

10     part-time.white                                       0.186**     0.000

11     part-time.black                                      -0.231**     0.001

     sex.color

12     male.white                                            0.059**     0.000

13     male.black                                           -0.083**     0.000

14   _cons                                                  14.294**     0.000

-------------------------------------------------------------------------------

*  p < .05

** p < .01

 

. poisgof

 

         Goodness-of-fit chi2  =  143357.3

         Prob > chi2(4)        =    0.0000

 

. poisgof, pearson

 

         Goodness-of-fit chi2  =  148750.5

         Prob > chi2(4)        =    0.0000

 

. *Standard errors are tiny, fit statistics are huge, coefficients differ a little bit from what we had before.

. *This second model is sort of a travesty, which is why clogg and eliason don't mention.

. desmat: poisson  wncount labor*sex=dev(3)*dev(2) labor*color=dev(3)*dev(3) sex*color=dev(2)*dev(3)

-------------------------------------------------------------------------------

   Poisson regression

-------------------------------------------------------------------------------

   Dependent variable                                                  wncount

   Optimization:                                                            ml

   Number of observations:                                                  18

   Initial log likelihood:                                          -79982.751

   Log likelihood:                                                    -130.057

   LR chi square:                                                   159705.387

   Model degrees of freedom:                                                13

   Pseudo R-squared:                                                     0.998

   Prob:                                                                 0.000

-------------------------------------------------------------------------------

nr Effect                                                    Coeff        s.e.

-------------------------------------------------------------------------------

   wncount

     labor

1      unemployed                                           -0.690**     0.020

2      part-time                                            -0.435**     0.018

     sex

3      male                                                  0.013       0.010

     labor.sex

4      unemployed.male                                       0.165**     0.009

5      part-time.male                                       -0.340**     0.007

     color

6      white                                                 1.858**     0.012

7      black                                                -0.133**     0.015

     labor.color

8      unemployed.white                                     -0.263**     0.020

9      unemployed.black                                      0.384**     0.023

10     part-time.white                                       0.186**     0.018

11     part-time.black                                      -0.232**     0.022

     sex.color

12     male.white                                            0.059**     0.009

13     male.black                                           -0.083**     0.011

14   _cons                                                   7.033**     0.012

-------------------------------------------------------------------------------

*  p < .05

** p < .01

 

. poisgof

 

         Goodness-of-fit chi2  =  100.2395

         Prob > chi2(4)        =    0.0000

 

. poisgof, pearson

 

         Goodness-of-fit chi2  =  104.0537

         Prob > chi2(4)        =    0.0000

 

. *These coefficients correspond exactly to clogg and eliason's second model. This is C+E's second model

. *The correct solution, in Clogg and Eliason's view, uses the weights to generate unbiased coefficients, but then uses the actual data to generate the SE of the coefficients

. *In stata, it requires using the 'exposure' option in poisson regression.

. *In stata, the exposure is the inverse of the sampling probability, which in this case is the inverse of the weights.

. gen inverse_weight=1/weight

 

. desmat: poisson uwcount labor*sex=dev(3)*dev(2) labor*color=dev(3)*dev(3) sex*color=dev(2)*dev(3), exposure(inverse_weight)

-------------------------------------------------------------------------------

   Poisson regression

-------------------------------------------------------------------------------

   Dependent variable                                                  uwcount

   Optimization:                                                            ml

   Number of observations:                                                  18

   Initial log likelihood:                                          -84619.027

   Log likelihood:                                                    -124.919

   LR chi square:                                                   168988.216

   Model degrees of freedom:                                                13

   Pseudo R-squared:                                                     0.999

   Prob:                                                                 0.000

-------------------------------------------------------------------------------

nr Effect                                                    Coeff        s.e.

-------------------------------------------------------------------------------

   uwcount

     labor

1      unemployed                                           -0.688**     0.018

2      part-time                                            -0.440**     0.017

     sex

3      male                                                  0.013       0.009

     labor.sex

4      unemployed.male                                       0.166**     0.009

5      part-time.male                                       -0.343**     0.007

     color

6      white                                                 1.860**     0.011

7      black                                                -0.136**     0.014

     labor.color

8      unemployed.white                                     -0.265**     0.018

9      unemployed.black                                      0.386**     0.022

10     part-time.white                                       0.191**     0.017

11     part-time.black                                      -0.238**     0.022

     sex.color

12     male.white                                            0.058**     0.009

13     male.black                                           -0.085**     0.011

14   _cons                                                  14.292**     0.011

     ln(inverse_weight)                                                (offset)

-------------------------------------------------------------------------------

*  p < .05

** p < .01

 

. poisgof

 

         Goodness-of-fit chi2  =  89.58815

         Prob > chi2(4)        =    0.0000

 

. poisgof, pearson

 

         Goodness-of-fit chi2  =  93.54537

         Prob > chi2(4)        =    0.0000

 

. *That's clogg and eliason model 3, my model 4

. exit, clear