class 17 on loglinear and nbreg

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

log: C:\AAA Miker Files\newer web pages\soc_388_notes\soc_388_2007\class_

> 17_log.log

log type: text

opened on: 27 Nov 2007, 11:34:18

. *A quick log about negative binomial and loglinear models.

. desmat: poisson count frace*fed6 mrace*med6 rendog blendog BW @BWsce

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

Poisson regression

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

Dependent variable count

Optimization: ml

Number of observations: 324

Initial log likelihood: -1335369.103

Log likelihood: -139712.934

LR chi square: 2391312.338

Model degrees of freedom: 38

Pseudo R-squared: 0.895

Prob: 0.000

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

nr Effect Coeff s.e.

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

count

frace

1 NH Black -3.412** 0.049

2 All Others -0.708** 0.017

fed6

3 10,11 0.342** 0.008

4 HS 2.069** 0.006

5 Some col 1.326** 0.007

6 BA 0.765** 0.007

7 >BA 0.115** 0.008

frace.fed6

8 NH Black.10,11 0.030 0.021

9 NH Black.HS -0.507** 0.017

10 NH Black.Some col -0.492** 0.019

11 NH Black.BA -0.998** 0.023

12 NH Black.>BA -0.977** 0.028

13 All Others.10,11 -0.361** 0.023

14 All Others.HS -1.029** 0.019

15 All Others.Some col -1.129** 0.022

16 All Others.BA -1.814** 0.031

17 All Others.>BA -1.541** 0.036

mrace

18 NH Black -1.959** 0.038

19 All Others -0.821** 0.017

med6

20 10,11 0.137** 0.008

21 HS 1.782** 0.006

22 Some col 1.223** 0.006

23 BA 0.739** 0.007

24 >BA 0.559** 0.007

mrace.med6

25 NH Black.10,11 0.034 0.019

26 NH Black.HS -0.584** 0.015

27 NH Black.Some col -0.721** 0.017

28 NH Black.BA -1.422** 0.023

29 NH Black.>BA -1.716** 0.026

30 All Others.10,11 -0.161** 0.024

31 All Others.HS -0.815** 0.019

32 All Others.Some col -0.840** 0.021

33 All Others.BA -1.584** 0.029

34 All Others.>BA -1.512** 0.030

rendog

35 1 2.291** 0.008

blendog

36 1 4.204** 0.073

37 1 -0.546** 0.039

38 BWsce 0.136** 0.012

39 _cons 5.226** 0.011

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

* p < .05

** p < .01

. poisgof

Goodness-of-fit chi2 = 277491.9

Prob > chi2(285) = 0.0000

. *This is the Model 1 loglinear model from my AJS paper. Note the positive and significant BWsce (Black white status caste exchange) term.

. desmat: nbreg count frace*fed6 mrace*med6 rendog blendog BW @BWsce, verbose

Desmat generated the following design matrix:

nr Variables Term Parameterization

First Last

1 _x_1 _x_2 frace ind(1)

2 _x_3 _x_7 fed6 ind(1)

3 _x_8 _x_17 frace.fed6 ind(1).ind(1)

4 _x_18 _x_19 mrace ind(1)

5 _x_20 _x_24 med6 ind(1)

6 _x_25 _x_34 mrace.med6 ind(1).ind(1)

7 _x_35 rendog ind(0)

8 _x_36 blendog ind(0)

9 _x_37 BW ind(0)

10 _x_38 BWsce direct

Fitting Poisson model:

Iteration 0: log likelihood = -1785213.3

Iteration 1: log likelihood = -1335702.9 (backed up)

Iteration 2: log likelihood = -1072685.9

Iteration 3: log likelihood = -420545.17

Iteration 4: log likelihood = -177722.64

Iteration 5: log likelihood = -140348.62

Iteration 6: log likelihood = -139713.77

Iteration 7: log likelihood = -139712.93

Iteration 8: log likelihood = -139712.93

Fitting constant-only model:

Iteration 0: log likelihood = -2750.3011

Iteration 1: log likelihood = -2216.8924

Iteration 2: log likelihood = -2216.8896

Iteration 3: log likelihood = -2216.8896

Fitting full model:

Iteration 0: log likelihood = -2154.7704 (not concave)

Iteration 1: log likelihood = -2034.0179

Iteration 2: log likelihood = -1992.6755

Iteration 3: log likelihood = -1941.2353

Iteration 4: log likelihood = -1853.3137

Iteration 5: log likelihood = -1852.3806

Iteration 6: log likelihood = -1852.3774

Iteration 7: log likelihood = -1852.3774

Negative binomial regression Number of obs = 324

LR chi2(38) = 729.02

Prob > chi2 = 0.0000

Log likelihood = -1852.3774 Pseudo R2 = 0.1644

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

count | Coef. Std. Err. z P>|z| [95% Conf. Interval]

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

_x_1 | -3.54226 .3671344 -9.65 0.000 -4.261831 -2.82269

_x_2 | -1.143256 .3275539 -3.49 0.000 -1.78525 -.5012625

_x_3 | .3411358 .3214777 1.06 0.289 -.2889489 .9712205

_x_4 | 1.617797 .3258008 4.97 0.000 .9792392 2.256355

_x_5 | 1.015074 .3359433 3.02 0.003 .3566369 1.67351

_x_6 | .4092661 .347168 1.18 0.238 -.2711706 1.089703

_x_7 | -.0381025 .3572391 -0.11 0.915 -.7382783 .6620733

_x_8 | .0191493 .4663474 0.04 0.967 -.8948748 .9331734

_x_9 | -.3327831 .4625601 -0.72 0.472 -1.239384 .573818

_x_10 | -.1818947 .467158 -0.39 0.697 -1.097508 .7337181

_x_11 | -.6255032 .4753208 -1.32 0.188 -1.557115 .3061085

_x_12 | -.5304081 .48833 -1.09 0.277 -1.487517 .4267012

_x_13 | -.3094955 .453883 -0.68 0.495 -1.19909 .5800989

_x_14 | -.6454152 .4537051 -1.42 0.155 -1.534661 .2438303

_x_15 | -.6475651 .4565164 -1.42 0.156 -1.542321 .2471907

_x_16 | -1.116904 .4595527 -2.43 0.015 -2.017611 -.2161974

_x_17 | -1.121978 .466346 -2.41 0.016 -2.036 -.2079567

_x_18 | -2.144572 .369716 -5.80 0.000 -2.869202 -1.419942

_x_19 | -.8657121 .3371297 -2.57 0.010 -1.526474 -.2049501

_x_20 | .0486761 .3283103 0.15 0.882 -.5948003 .6921525

_x_21 | 1.364141 .3305427 4.13 0.000 .7162892 2.011993

_x_22 | 1.025122 .3393403 3.02 0.003 .3600274 1.690217

_x_23 | .7357188 .3496288 2.10 0.035 .0504589 1.420979

_x_24 | 1.098691 .3597084 3.05 0.002 .3936753 1.803706

_x_25 | .1557508 .4615805 0.34 0.736 -.7489303 1.060432

_x_26 | -.2930492 .4610576 -0.64 0.525 -1.196706 .6106071

_x_27 | -.2825236 .4639913 -0.61 0.543 -1.19193 .6268826

_x_28 | -.9211377 .4720356 -1.95 0.051 -1.84631 .0040351

_x_29 | -1.275697 .481366 -2.65 0.008 -2.219157 -.3322372

_x_30 | -.1053643 .4646886 -0.23 0.821 -1.016137 .8054086

_x_31 | -.6205033 .4618318 -1.34 0.179 -1.525677 .2846704

_x_32 | -.6304821 .464632 -1.36 0.175 -1.541144 .2801799

_x_33 | -1.337587 .4705215 -2.84 0.004 -2.259792 -.4153817

_x_34 | -1.439732 .4750442 -3.03 0.002 -2.370801 -.5086626

_x_35 | 2.276199 .1601957 14.21 0.000 1.962222 2.590177

_x_36 | 4.065725 .4047264 10.05 0.000 3.272475 4.858974

_x_37 | -.6528239 .2048047 -3.19 0.001 -1.054234 -.2514141

_x_38 | .0450504 .083835 0.54 0.591 -.1192633 .2093641

_cons | 5.599588 .3222667 17.38 0.000 4.967957 6.231219

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

/lnalpha | -.1060076 .0771399 -.2571991 .0451839

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

alpha | .8994178 .069381 .7732142 1.04622

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

Likelihood-ratio test of alpha=0: chibar2(01) = 2.8e+05 Prob>=chibar2 = 0.000

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

Negative binomial regression

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

Dependent variable count

Optimization: ml

Number of observations: 324

Initial log likelihood: -2216.890

Log likelihood: -1852.377

LR chi square: 729.025

Model degrees of freedom: 38

Pseudo R-squared: 0.164

Dispersion: mean

Prob: 0.000

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

nr Effect Coeff s.e.

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

count

frace

1 NH Black -3.542** 0.367

2 All Others -1.143** 0.328

fed6

3 10,11 0.341 0.321

4 HS 1.618** 0.326

5 Some col 1.015** 0.336

6 BA 0.409 0.347

7 >BA -0.038 0.357

frace.fed6

8 NH Black.10,11 0.019 0.466

9 NH Black.HS -0.333 0.463

10 NH Black.Some col -0.182 0.467

11 NH Black.BA -0.626 0.475

12 NH Black.>BA -0.530 0.488

13 All Others.10,11 -0.309 0.454

14 All Others.HS -0.645 0.454

15 All Others.Some col -0.648 0.457

16 All Others.BA -1.117* 0.460

17 All Others.>BA -1.122* 0.466

mrace

18 NH Black -2.145** 0.370

19 All Others -0.866* 0.337

med6

20 10,11 0.049 0.328

21 HS 1.364** 0.331

22 Some col 1.025** 0.339

23 BA 0.736* 0.350

24 >BA 1.099** 0.360

mrace.med6

25 NH Black.10,11 0.156 0.462

26 NH Black.HS -0.293 0.461

27 NH Black.Some col -0.283 0.464

28 NH Black.BA -0.921 0.472

29 NH Black.>BA -1.276** 0.481

30 All Others.10,11 -0.105 0.465

31 All Others.HS -0.621 0.462

32 All Others.Some col -0.630 0.465

33 All Others.BA -1.338** 0.471

34 All Others.>BA -1.440** 0.475

rendog

35 1 2.276** 0.160

blendog

36 1 4.066** 0.405

37 1 -0.653** 0.205

38 BWsce 0.045 0.084

39 _cons 5.600** 0.322

lnalpha

40 _cons -0.106 0.077

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

* p < .05

** p < .01

. *The negative binomial version of M1 totally erases what was a significant and positive finding for the status-caste exchange variable.

. *And nbreg fits dramatically better, adding one additional term.

. desmat: poisson count frace*fed6 mrace*med6 med6*fed6 rendog blendog BW @B

> Wsce

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

Poisson regression

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

Dependent variable count

Optimization: ml

Number of observations: 324

Initial log likelihood: -1335369.103

Log likelihood: -2153.154

LR chi square: 2666431.898

Model degrees of freedom: 63

Pseudo R-squared: 0.998

Prob: 0.000

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

nr Effect Coeff s.e.

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

count

frace

1 NH Black -3.868** 0.050

2 All Others -0.931** 0.018

fed6

3 10,11 -0.441** 0.014

4 HS 0.237** 0.012

5 Some col -1.613** 0.021

6 BA -3.226** 0.046

7 >BA -3.938** 0.063

frace.fed6

8 NH Black.10,11 0.117** 0.021

9 NH Black.HS -0.204** 0.018

10 NH Black.Some col 0.063** 0.021

11 NH Black.BA -0.110** 0.026

12 NH Black.>BA 0.063* 0.031

13 All Others.10,11 -0.290** 0.023

14 All Others.HS -0.835** 0.020

15 All Others.Some col -0.835** 0.023

16 All Others.BA -1.387** 0.032

17 All Others.>BA -1.085** 0.037

mrace

18 NH Black -2.055** 0.038

19 All Others -1.057** 0.018

med6

20 10,11 -0.592** 0.015

21 HS -0.021 0.013

22 Some col -1.607** 0.022

23 BA -3.300** 0.049

24 >BA -3.669** 0.059

mrace.med6

25 NH Black.10,11 0.046* 0.019

26 NH Black.HS -0.524** 0.016

27 NH Black.Some col -0.700** 0.019

28 NH Black.BA -1.405** 0.025

29 NH Black.>BA -1.723** 0.029

30 All Others.10,11 -0.076** 0.024

31 All Others.HS -0.601** 0.020

32 All Others.Some col -0.554** 0.022

33 All Others.BA -1.202** 0.030

34 All Others.>BA -1.105** 0.032

med6.fed6

35 10,11.10,11 0.927** 0.020

36 10,11.HS 0.946** 0.017

37 10,11.Some col 1.067** 0.028

38 10,11.BA 0.800** 0.062

39 10,11.>BA 0.623** 0.089

40 HS.10,11 1.085** 0.017

41 HS.HS 2.298** 0.015

42 HS.Some col 2.680** 0.024

43 HS.BA 2.868** 0.048

44 HS.>BA 2.506** 0.066

45 Some col.10,11 1.232** 0.028

46 Some col.HS 2.904** 0.023

47 Some col.Some col 4.582** 0.029

48 Some col.BA 5.043** 0.050

49 Some col.>BA 4.763** 0.067

50 BA.10,11 1.002** 0.063

51 BA.HS 3.493** 0.051

52 BA.Some col 5.626** 0.053

53 BA.BA 7.368** 0.067

54 BA.>BA 6.899** 0.080

55 >BA.10,11 0.615** 0.081

56 >BA.HS 3.081** 0.061

57 >BA.Some col 5.576** 0.063

58 >BA.BA 7.465** 0.075

59 >BA.>BA 8.163** 0.086

rendog

60 1 2.261** 0.008

blendog

61 1 4.320** 0.073

62 1 -0.541** 0.039

63 BWsce 0.070** 0.018

64 _cons 6.949** 0.012

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

* p < .05

** p < .01

. poisgof

Goodness-of-fit chi2 = 2372.308

Prob > chi2(260) = 0.0000

That’s loglinear model 2.

. desmat: nbreg count frace*fed6 mrace*med6 med6*fed6 rendog blendog BW @BWsce

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

Negative binomial regression

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

Dependent variable count

Optimization: ml

Number of observations: 324

Initial log likelihood: -2216.890

Log likelihood: -1427.298

LR chi square: 1579.182

Model degrees of freedom: 63

Pseudo R-squared: 0.356

Dispersion: mean

Prob: 0.000

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

nr Effect Coeff s.e.

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

count

frace

1 NH Black -3.627** 0.116

2 All Others -1.021** 0.084

fed6

3 10,11 -0.313* 0.126

4 HS 0.312* 0.123

5 Some col -1.243** 0.133

6 BA -2.929** 0.152

7 >BA -3.412** 0.166

frace.fed6

8 NH Black.10,11 0.010 0.134

9 NH Black.HS -0.358** 0.126

10 NH Black.Some col -0.105 0.128

11 NH Black.BA -0.317* 0.134

12 NH Black.>BA -0.255 0.140

13 All Others.10,11 -0.341** 0.116

14 All Others.HS -0.749** 0.110

15 All Others.Some col -0.745** 0.112

16 All Others.BA -1.194** 0.120

17 All Others.>BA -1.120** 0.126

mrace

18 NH Black -1.981** 0.105

19 All Others -0.972** 0.090

med6

20 10,11 -0.486** 0.130

21 HS 0.175 0.127

22 Some col -1.141** 0.136

23 BA -2.872** 0.159

24 >BA -3.120** 0.169

mrace.med6

25 NH Black.10,11 0.043 0.123

26 NH Black.HS -0.467** 0.117

27 NH Black.Some col -0.516** 0.120

28 NH Black.BA -1.121** 0.127

29 NH Black.>BA -1.338** 0.133

30 All Others.10,11 -0.147 0.125

31 All Others.HS -0.585** 0.118

32 All Others.Some col -0.542** 0.119

33 All Others.BA -1.178** 0.125

34 All Others.>BA -1.154** 0.127

med6.fed6

35 10,11.10,11 0.924** 0.154

36 10,11.HS 0.998** 0.152

37 10,11.Some col 1.048** 0.165

38 10,11.BA 0.853** 0.197

39 10,11.>BA 0.302 0.218

40 HS.10,11 1.067** 0.152

41 HS.HS 2.150** 0.148

42 HS.Some col 2.370** 0.158

43 HS.BA 2.595** 0.181

44 HS.>BA 2.012** 0.195

45 Some col.10,11 1.149** 0.164

46 Some col.HS 2.587** 0.157

47 Some col.Some col 3.997** 0.165

48 Some col.BA 4.419** 0.185

49 Some col.>BA 4.168** 0.198

50 BA.10,11 1.060** 0.202

51 BA.HS 3.225** 0.183

52 BA.Some col 5.027** 0.189

53 BA.BA 6.695** 0.203

54 BA.>BA 6.210** 0.216

55 >BA.10,11 0.492* 0.220

56 >BA.HS 2.830** 0.194

57 >BA.Some col 4.914** 0.198

58 >BA.BA 6.829** 0.212

59 >BA.>BA 7.477** 0.221

rendog

60 1 2.202** 0.037

blendog

61 1 4.162** 0.125

62 1 -0.608** 0.067

63 BWsce 0.015 0.025

64 _cons 6.708** 0.105

lnalpha

65 _cons -3.250** 0.122

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

* p < .05

** p < .01

* And that’s the nbreg version of model 2, which again erases the significance of the BWsce term.