Hi guys,
I am estimating a model in willingness-to-pay space with mixlogitwtp . First,I really like how straightforward mixlogitwtp and mixlogit are, and they are a big help to my research.
However I encounter issue with mixlogitwtp when halton draws is at 2000, specifically convergence issues. I wonder if anyone runs into similar issues when halton gets high.
After a few rounds of research, we have tried a couple of suggestions, especially from Hong (https://www.statalist.org/forums/for...it#post1545090). As my preference space model has no problem at halton 2000, i wonder if something I can incorporate in the code to make it work. I will update the data sample as soon as my current stata program stops running.
codes:
mixlogitwtp selected, group(_caseid) id(res_id) price (mprice) rand($X) nrep(500) ////[can converge]
mat b=e(b)
mat v=e(v)
this is the code that I have problem with:
mixlogitwtp selected, group(_caseid) id(res_id) price (mprice) rand($X) nrep(2000) technique(bfgs 5 nr 5) from (b)
---
(setting technique to bfgs)
Iteration 0: Log likelihood = -4159.8497
Iteration 1: Log likelihood = -4156.885 (backed up)
Iteration 2: Log likelihood = -4156.123 (backed up)
Iteration 3: Log likelihood = -4140.8442
Iteration 4: Log likelihood = -4138.2942
(switching technique to nr)
Iteration 5: Log likelihood = -4135.0974
Iteration 6: Log likelihood = -4133.2778 (not concave)
Iteration 7: Log likelihood = -4128.9887 (not concave)
Iteration 8: Log likelihood = -4124.2353
Iteration 9: Log likelihood = -4122.9876 (not concave)
(switching technique to bfgs)
Iteration 10: Log likelihood = -4122.2715
Iteration 11: Log likelihood = -4121.4353
Iteration 12: Log likelihood = -4121.1275
Iteration 13: Log likelihood = -4120.8714
Iteration 14: Log likelihood = -4120.3652
(switching technique to nr)
Iteration 15: Log likelihood = -4120.2519 (not concave)
Iteration 16: Log likelihood = -4119.9666 (not concave)
Iteration 17: Log likelihood = -4119.9068 (not concave)
Iteration 18: Log likelihood = -4119.884 (not concave)
Iteration 19: Log likelihood = -4119.8739 (not concave)
(switching technique to bfgs)
Iteration 20: Log likelihood = -4119.8608
Iteration 21: Log likelihood = -4119.8606
Iteration 22: Log likelihood = -4119.8518
Iteration 23: Log likelihood = -4119.8445
Iteration 24: Log likelihood = -4119.8322
(switching technique to nr)
Iteration 25: Log likelihood = -4119.8321 (not concave)
Iteration 26: Log likelihood = -4119.8321 (not concave)
Iteration 27: Log likelihood = -4119.8321 (not concave)
Iteration 28: Log likelihood = -4119.8321 (not concave)
Iteration 29: Log likelihood = -4119.8321 (not concave)
(switching technique to bfgs)
Iteration 30: Log likelihood = -4119.8321 (backed up)
Iteration 31: Log likelihood = -4119.8316
Iteration 32: Log likelihood = -4119.827
Iteration 33: Log likelihood = -4119.8263
Iteration 34: Log likelihood = -4119.8259
(switching technique to nr)
Iteration 35: Log likelihood = -4119.8256 (not concave)
Iteration 36: Log likelihood = -4119.8227 (not concave)
Iteration 37: Log likelihood = -4119.8226 (not concave)
Iteration 38: Log likelihood = -4119.8221 (not concave)
Iteration 39: Log likelihood = -4119.8212 (not concave)
(switching technique to bfgs)
Iteration 40: Log likelihood = -4119.8188
Iteration 41: Log likelihood = -4119.8123
Iteration 42: Log likelihood = -4119.8055
Iteration 43: Log likelihood = -4119.7964
Iteration 44: Log likelihood = -4119.7879
(switching technique to nr)
Iteration 45: Log likelihood = -4119.7852 (not concave)
Iteration 46: Log likelihood = -4119.7835 (not concave)
Iteration 47: Log likelihood = -4119.7768
Iteration 48: Log likelihood = -4119.7712 (not concave)
Iteration 49: Log likelihood = -4119.7667 (not concave)
(switching technique to bfgs)
Iteration 50: Log likelihood = -4119.7592
Iteration 51: Log likelihood = -4119.7553
Iteration 52: Log likelihood = -4119.7525
Iteration 53: Log likelihood = -4119.751
Iteration 54: Log likelihood = -4119.7495
(switching technique to nr)
Iteration 55: Log likelihood = -4119.7489 (not concave)
Iteration 56: Log likelihood = -4119.7473 (not concave)
Iteration 57: Log likelihood = -4119.7472 (not concave)
Iteration 58: Log likelihood = -4119.7471 (not concave)
Iteration 59: Log likelihood = -4119.747 (not concave)
(switching technique to bfgs)
Iteration 60: Log likelihood = -4119.7469
Iteration 61: Log likelihood = -4119.7468
Iteration 62: Log likelihood = -4119.7463
Iteration 63: Log likelihood = -4119.746
Iteration 64: Log likelihood = -4119.746
(switching technique to nr)
Iteration 65: Log likelihood = -4119.7459 (not concave)
Iteration 66: Log likelihood = -4119.7459 (not concave)
Iteration 67: Log likelihood = -4119.7458 (not concave)
Iteration 68: Log likelihood = -4119.7458 (not concave)
Iteration 69: Log likelihood = -4119.7458 (not concave)
(switching technique to bfgs)
Iteration 70: Log likelihood = -4119.7457
Iteration 71: Log likelihood = -4119.7457 (backed up)
Iteration 72: Log likelihood = -4119.7457 (backed up)
Iteration 73: Log likelihood = -4119.7457 (backed up)
Iteration 74: Log likelihood = -4119.7457 (backed up)
(switching technique to nr)
Iteration 75: Log likelihood = -4119.7457 (not concave)
Iteration 76: Log likelihood = -4119.7457 (not concave)
Iteration 77: Log likelihood = -4119.7457 (not concave)
Iteration 78: Log likelihood = -4119.7457 (not concave)
Iteration 79: Log likelihood = -4119.7457 (not concave)
(switching technique to bfgs)
Iteration 80: Log likelihood = -4119.7457
I am estimating a model in willingness-to-pay space with mixlogitwtp . First,I really like how straightforward mixlogitwtp and mixlogit are, and they are a big help to my research.
However I encounter issue with mixlogitwtp when halton draws is at 2000, specifically convergence issues. I wonder if anyone runs into similar issues when halton gets high.
After a few rounds of research, we have tried a couple of suggestions, especially from Hong (https://www.statalist.org/forums/for...it#post1545090). As my preference space model has no problem at halton 2000, i wonder if something I can incorporate in the code to make it work. I will update the data sample as soon as my current stata program stops running.
codes:
mixlogitwtp selected, group(_caseid) id(res_id) price (mprice) rand($X) nrep(500) ////[can converge]
mat b=e(b)
mat v=e(v)
this is the code that I have problem with:
mixlogitwtp selected, group(_caseid) id(res_id) price (mprice) rand($X) nrep(2000) technique(bfgs 5 nr 5) from (b)
---
(setting technique to bfgs)
Iteration 0: Log likelihood = -4159.8497
Iteration 1: Log likelihood = -4156.885 (backed up)
Iteration 2: Log likelihood = -4156.123 (backed up)
Iteration 3: Log likelihood = -4140.8442
Iteration 4: Log likelihood = -4138.2942
(switching technique to nr)
Iteration 5: Log likelihood = -4135.0974
Iteration 6: Log likelihood = -4133.2778 (not concave)
Iteration 7: Log likelihood = -4128.9887 (not concave)
Iteration 8: Log likelihood = -4124.2353
Iteration 9: Log likelihood = -4122.9876 (not concave)
(switching technique to bfgs)
Iteration 10: Log likelihood = -4122.2715
Iteration 11: Log likelihood = -4121.4353
Iteration 12: Log likelihood = -4121.1275
Iteration 13: Log likelihood = -4120.8714
Iteration 14: Log likelihood = -4120.3652
(switching technique to nr)
Iteration 15: Log likelihood = -4120.2519 (not concave)
Iteration 16: Log likelihood = -4119.9666 (not concave)
Iteration 17: Log likelihood = -4119.9068 (not concave)
Iteration 18: Log likelihood = -4119.884 (not concave)
Iteration 19: Log likelihood = -4119.8739 (not concave)
(switching technique to bfgs)
Iteration 20: Log likelihood = -4119.8608
Iteration 21: Log likelihood = -4119.8606
Iteration 22: Log likelihood = -4119.8518
Iteration 23: Log likelihood = -4119.8445
Iteration 24: Log likelihood = -4119.8322
(switching technique to nr)
Iteration 25: Log likelihood = -4119.8321 (not concave)
Iteration 26: Log likelihood = -4119.8321 (not concave)
Iteration 27: Log likelihood = -4119.8321 (not concave)
Iteration 28: Log likelihood = -4119.8321 (not concave)
Iteration 29: Log likelihood = -4119.8321 (not concave)
(switching technique to bfgs)
Iteration 30: Log likelihood = -4119.8321 (backed up)
Iteration 31: Log likelihood = -4119.8316
Iteration 32: Log likelihood = -4119.827
Iteration 33: Log likelihood = -4119.8263
Iteration 34: Log likelihood = -4119.8259
(switching technique to nr)
Iteration 35: Log likelihood = -4119.8256 (not concave)
Iteration 36: Log likelihood = -4119.8227 (not concave)
Iteration 37: Log likelihood = -4119.8226 (not concave)
Iteration 38: Log likelihood = -4119.8221 (not concave)
Iteration 39: Log likelihood = -4119.8212 (not concave)
(switching technique to bfgs)
Iteration 40: Log likelihood = -4119.8188
Iteration 41: Log likelihood = -4119.8123
Iteration 42: Log likelihood = -4119.8055
Iteration 43: Log likelihood = -4119.7964
Iteration 44: Log likelihood = -4119.7879
(switching technique to nr)
Iteration 45: Log likelihood = -4119.7852 (not concave)
Iteration 46: Log likelihood = -4119.7835 (not concave)
Iteration 47: Log likelihood = -4119.7768
Iteration 48: Log likelihood = -4119.7712 (not concave)
Iteration 49: Log likelihood = -4119.7667 (not concave)
(switching technique to bfgs)
Iteration 50: Log likelihood = -4119.7592
Iteration 51: Log likelihood = -4119.7553
Iteration 52: Log likelihood = -4119.7525
Iteration 53: Log likelihood = -4119.751
Iteration 54: Log likelihood = -4119.7495
(switching technique to nr)
Iteration 55: Log likelihood = -4119.7489 (not concave)
Iteration 56: Log likelihood = -4119.7473 (not concave)
Iteration 57: Log likelihood = -4119.7472 (not concave)
Iteration 58: Log likelihood = -4119.7471 (not concave)
Iteration 59: Log likelihood = -4119.747 (not concave)
(switching technique to bfgs)
Iteration 60: Log likelihood = -4119.7469
Iteration 61: Log likelihood = -4119.7468
Iteration 62: Log likelihood = -4119.7463
Iteration 63: Log likelihood = -4119.746
Iteration 64: Log likelihood = -4119.746
(switching technique to nr)
Iteration 65: Log likelihood = -4119.7459 (not concave)
Iteration 66: Log likelihood = -4119.7459 (not concave)
Iteration 67: Log likelihood = -4119.7458 (not concave)
Iteration 68: Log likelihood = -4119.7458 (not concave)
Iteration 69: Log likelihood = -4119.7458 (not concave)
(switching technique to bfgs)
Iteration 70: Log likelihood = -4119.7457
Iteration 71: Log likelihood = -4119.7457 (backed up)
Iteration 72: Log likelihood = -4119.7457 (backed up)
Iteration 73: Log likelihood = -4119.7457 (backed up)
Iteration 74: Log likelihood = -4119.7457 (backed up)
(switching technique to nr)
Iteration 75: Log likelihood = -4119.7457 (not concave)
Iteration 76: Log likelihood = -4119.7457 (not concave)
Iteration 77: Log likelihood = -4119.7457 (not concave)
Iteration 78: Log likelihood = -4119.7457 (not concave)
Iteration 79: Log likelihood = -4119.7457 (not concave)
(switching technique to bfgs)
Iteration 80: Log likelihood = -4119.7457
Comment