st: Stgest command returning unexpected results

[Chelsea Erickson <chelseae@email.unc.edu>]

Dear Stata Users,

I am using the stgest command from The Stata Journal Vol. 2 No. 2 
(2002) to implement G-estimation in Stata 10.

Using my data, I have a reasonable range of values for the psi estimate 
that correspond to a z value of 0.  (I found this range using a grid 
search.) However, the bisection search algorithm does not work properly 
in identifying the point estimate or the confidence interval for psi 
because smaller values of psi correspond to smaller z-values and 
accordingly, larger values of psi correspond to larger z-values.  This 
trend works in the opposite direction than that suggested by the 
algorithm in Sterne and Tilling's example in the original article.

I'm wondering what these results mean or if there's potential for more 
than one solution to the estimate of psi?

I've pasted results I've gotten so far using the grid search option.  
(Note: The confidence limits below are reversed because of this 
unexpected trend.)


Thank you,
Chelsea Erickson


    +-----------------------------+
    |  psi       pval           z |
    |-----------------------------|
 1. |   -2   .1536383   -1.426797 |
 2. | -1.9   .1533419   -1.427826 |
 3. | -1.8   .1529408   -1.429221 |
 4. | -1.7   .1524907   -1.430789 |
 5. | -1.6   .1506994   -1.437066 |
    |-----------------------------|
 6. | -1.5   .1509089   -1.436329 |
 7. | -1.4   .1513166   -1.434896 |
 8. | -1.3   .1521363   -1.432026 |
 9. | -1.2   .1536152   -1.426877 |
10. | -1.1   .1560282   -1.418557 |
    |-----------------------------|
11. |   -1   .1596726   -1.406174 |
12. |  -.9   .1648646   -1.388895 |
13. |  -.8    .171939      -1.366 |
14. |  -.7   .1812494   -1.336918 |
15. |  -.6   .1921174   -1.304341 |
    |-----------------------------|
16. |  -.5   .2075079   -1.260448 |
17. |  -.4   .2261214    -1.21041 |
18. |  -.3   .2484346    -1.15416 |
19. |  -.2   .2748978   -1.091853 |
20. |  -.1   .3059151   -1.023831 |
    |-----------------------------|
21. |    0   .3418189   -.9505774 |
22. |   .1   .3828435   -.8726691 |
23. |   .2   .4290949   -.7907415 |
24. |   .3   .4805235   -.7054608 |
25. |   .4   .5369009   -.6175062 |
    |-----------------------------|
26. |   .5   .5978044   -.5275605 |
27. |   .6    .662615   -.4363057 |
28. |   .7   .7305279   -.3444234 |
29. |   .8    .800581   -.2525952 |
30. |   .9   .8716983   -.1615017 |
    |-----------------------------|
31. |    1   .9427495   -.0718146 |
32. |  1.1   .9873754    .0158232 |
33. |  1.2   .9196773    .1008402 |
34. |  1.3   .8549415    .1828171 |
35. |  1.4   .7936096    .2616262 |
    |-----------------------------|
36. |  1.5   .7356226    .3376558 |
37. |  1.6   .6802456    .4121279 |
38. |  1.7     .62593    .4874634 |
39. |  1.8   .5703663    .5675122 |
40. |  1.9   .5109847    .6573053 |
    |-----------------------------|
41. |    2   .4460377    .7620375 |
42. |  2.1   .3758833    .8855069 |
43. |  2.2   .3035411    1.028869 |
44. |  2.3   .2338706    1.190447 |
45. |  2.4   .1717895    1.366476 |
    |-----------------------------|
46. |  2.5   .1206355    1.552112 |
47. |  2.6    .081475    1.742191 |
48. |  2.7   .0534084    1.931604 |
49. |  2.8   .0344012    2.115337 |
50. |  2.9   .0221206    2.288291 |
    |-----------------------------|
51. |    3   .0144889    2.444909 |
52. |  3.1   .0099185    2.578656 |
53. |  3.2   .0073333    2.681317 |
54. |  3.3   .0061042     2.74213 |
55. |  3.4   .0060161    2.746904 |
    |-----------------------------|
56. |  3.5   .0074167    2.677532 |
    +-----------------------------+

G estimate of psi for educ: 1.100 (95% CI 2.700 to -1.600)








*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/