www.cmm.bristol.ac.uk/forum

Hi,

I try to run a discrete-time multinomial event history model with four exit states (work, active labour market participation, social security benefit entitlement and other) for 13,552 social assistance beneficiaries for their repeated spells in social assistance. I have a stepwise specification of the baseline hazard function and the spell number is included as a covariate (3 categories). Furthermore, I try to assess the association of various time-varying beneficiary characteristics with the socio-economic state after exit from social assistance. I use the igls-estimations as priors for the mcmc-estimation.

When I included a random intercept in the model with stata 12 that invokes MLwiN v2.32, I get the following error message:
“error while obeying batch file C:\Users\scarpent\AppData\Local\Temp\ST_02000007.tmp at line number 180:
calc c1177 = expo('h')
38756 numeric errors in calculate command, first 20 affected entries listed.
Affected entries replaced with system missing.”

My model is the following:
sort uniekid
xi: runmlwin Yvar_rsp cons i.stepwise_yvarrspseq_5cat i.Yvar_rsp_spell_cat3 i.kalenderjaarkwartaal i.geslacht_pg c_leeftijd_t1 c_lft_t1_squared i.householdtype_noc_other ///
i.cat_gebplaatsoud7_ad i.voork_sancrva_lastyear c_pc_werkend_vorigjaar c_pc_szh_vorigjaar i.student_obvllrechtgk i.anderenatio_tv i.kandidaatvluchteling_ad i.wetg_nomenc2zondernul if Yvar_rsp_seq > 1, ///
level2(insz: cons) level1(uniekid) ///
discrete(distribution(multinomial) link(mlogit) den(cons) basecategory(0)) maxiterations(100) nopause

sort uniekid
xi: runmlwin Yvar_rsp cons i.stepwise_yvarrspseq_5cat i.Yvar_rsp_spell_cat3 i.kalenderjaarkwartaal i.geslacht_pg c_leeftijd_t1 c_lft_t1_squared i.householdtype_noc_other ///
i.cat_gebplaatsoud7_ad i.voork_sancrva_lastyear c_pc_werkend_vorigjaar c_pc_szh_vorigjaar i.student_obvllrechtgk i.anderenatio_tv i.kandidaatvluchteling_ad i.wetg_nomenc2zondernul if Yvar_rsp_seq > 1, ///
level2 (insz: cons, residuals(all2)) level1 (uniekid) ///
discrete(distribution(multinomial) link(mlogit) den(cons) basecategory(0)) mcmc(burnin(4000) chain(40000) orthogonal) initsprevious nopause saveworksheet(saveworksheet_all2)

It is not clear to me how I should proceed.
Can you help me further?

Best,

S

Hi Sarah,

The IGLS model which you are using to provide starting values for your MCMC model appears to be falling over

Two suggestions.

(1) Fit a single-level lGLS model instead and then manually specify values for the 3*3 level-2 covariance matrix (only set the variances, set all covariances to zero for simplicity)

(2) Don't fit an IGLS model at all, but manually specify starting values for all model parameters

I hope this helps

George