Hello -
I have multilevel data - births nested within mothers who were interviewed in 3 panels so giving information on new births in each panel - some are only in 1 panel, some in 2, and some in all 3. I have constructed birth histories for them using the 3 panels. I am running models of singleton births in multiparous mothers. I am trying to run random effect models to see what the effect of prior stillbirth is on a future stillbirth. I have xtset the data using the mother's id (pidlink_n). I ran my RE model using xtlogit (but did not use sampling weights) and got a significant result re: effect of prior stillbirth. When I run the same model using melogit, I get similar results. However, when I run the model using melogit and use sampling weights at level 1 (even through these are sampling weights at level 2 - i.e. it is the probability of the mother being in the sample but I am assuming that equals the probability of her births being in the sample), I get vastly different results. I am not sure if I am weighting correctly and which result is correct. I was hoping someone can help me figure out what is the right approach here and whether weighting should be done. See code and output below and thanks in advance!
*Just random effects without weighting
xtlogit stillbirth stillbirth_hx ib2.parity_grp urban ib1.educlevel age2 age_mom_birth ib3.panel if in_analysis == 1 & max_parity > 0 & twin != 1, vce(robust) or
*Mixed effects without weighting
melogit stillbirth stillbirth_hx ib2.parity_grp urban ib1.educlevel age2 age_mom_birth ib3.panel if in_analysis == 1 & max_parity > 0 & twin != 1 || pidlink: , or
*Mixed effects with weighting
melogit stillbirth stillbirth_hx ib2.parity_grp urban ib1.educlevel age2 age_mom_birth ib3.panel if in_analysis == 1 & max_parity > 0 & twin != 1 [pw=wgt] || pidlink: , or
I have multilevel data - births nested within mothers who were interviewed in 3 panels so giving information on new births in each panel - some are only in 1 panel, some in 2, and some in all 3. I have constructed birth histories for them using the 3 panels. I am running models of singleton births in multiparous mothers. I am trying to run random effect models to see what the effect of prior stillbirth is on a future stillbirth. I have xtset the data using the mother's id (pidlink_n). I ran my RE model using xtlogit (but did not use sampling weights) and got a significant result re: effect of prior stillbirth. When I run the same model using melogit, I get similar results. However, when I run the model using melogit and use sampling weights at level 1 (even through these are sampling weights at level 2 - i.e. it is the probability of the mother being in the sample but I am assuming that equals the probability of her births being in the sample), I get vastly different results. I am not sure if I am weighting correctly and which result is correct. I was hoping someone can help me figure out what is the right approach here and whether weighting should be done. See code and output below and thanks in advance!
*Just random effects without weighting
xtlogit stillbirth stillbirth_hx ib2.parity_grp urban ib1.educlevel age2 age_mom_birth ib3.panel if in_analysis == 1 & max_parity > 0 & twin != 1, vce(robust) or
| (Std. Err. adjusted | for 5,002 | clusters in pidlink_n) | ||
| Robust | ||||
| stillbirth | Odds Ratio | Std. Err. z | P>z | [95% Conf. Interval] |
| stillbirth_hx | 6.460225 | 2.888999 4.17 | 0.000 | 2.689008 15.52041 |
| parity_grp | ||||
| none | 1.520591 | .2725703 2.34 | 0.019 | 1.070119 2.160691 |
| 3 or more | .8054398 | .2188812 -0.80 | 0.426 | .472842 1.371988 |
*Mixed effects without weighting
melogit stillbirth stillbirth_hx ib2.parity_grp urban ib1.educlevel age2 age_mom_birth ib3.panel if in_analysis == 1 & max_parity > 0 & twin != 1 || pidlink: , or
| stillbirth | Odds Ratio | Std. Err. | z | P>z | [95% Conf. | Interval] |
| stillbirth_hx | 6.488712 | 1.36007 | 8.92 | 0.000 | 4.302708 | 9.785321 |
| parity_grp | ||||||
| none | 1.520723 | .2591677 | 2.46 | 0.014 | 1.088893 | 2.123809 |
| 3 or more | .8053737 | .2014848 | -0.87 | 0.387 | .4932279 | 1.315065 |
*Mixed effects with weighting
melogit stillbirth stillbirth_hx ib2.parity_grp urban ib1.educlevel age2 age_mom_birth ib3.panel if in_analysis == 1 & max_parity > 0 & twin != 1 [pw=wgt] || pidlink: , or
| Robust | ||||||
| stillbirth | Odds Ratio | Std. Err. | z | P>z | [95% Conf. | Interval] |
| stillbirth_hx | 1.083401 | .6682667 | 0.13 | 0.897 | .3234062 | 3.62936 |
| parity_grp | ||||||
| none | 1.251625 | .2735086 | 1.03 | 0.304 | .8155812 | 1.920796 |
| 3 or more | 1.005602 | .3621235 | 0.02 | 0.988 | .4964818 | 2.036803 |
