Dear All,
I am interested to examine the effect of IPV (Intimate Partner Violence) on my proxy outcome measurement for skilled maternal health care utilisation:

Dependent Variable - adequate ANC visits (1) Vs. Inadequate ANC visits (0); and
  • Health Facility Delivery (1) Vs. Home delivery (0)); whether it differ by mother education and household wealth status:
Main exposures: spousal emotional IPV (Yes/No or 1/0) and spousal Physical IPV (Yes/No or 1/0).

Stratified by: (Education status) - Lower education (1) & Higher education (2).

All other my covariates are also categorical variables. I have fitted multilevel logistic regression because the data is clustered at the survey level. My setting is .Windows 10 Stata 16.1.

Model I. Education stratified adjusted logistic regression
Output:

. melogit anc_adequacy i.emo_Ipv i.age_catgorey i.husband_educ i.wealth_hh i.mediae_expo i.birth_order i.dma i.V102 i.contextual_regio
> ns if educ_mom==1 || psu :,or nolog

Mixed-effects logistic regression Number of obs = 2,548
Group variable: psu Number of groups = 580

Obs per group:
min = 1
avg = 4.4
max = 12

Integration method: mvaghermite Integration pts. = 7

Wald chi2(15) = 160.53
Log likelihood = -1384.4603 Prob > chi2 = 0.0000
------------------------------------------------------------------------------------------------
anc_adequacy | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------------------------+----------------------------------------------------------------
emo_Ipv |
Yes | 1.353753 .1777473 2.31 0.021 1.046591 1.751064
|

In the above model, the main effect of emotional violence (emo_ipv) on ANC (adequate ANC), is itself quite weak (AOR = 1.35). the odds ratio becomes greater than 1, and we will see that the lower limit of the odds ratio would be close to 1. What statistical consideration should be here for this weak relationship?


Model II.

Output:

melogit del_place i.phy_ipv i.age_catgorey i.husband_educ i.wealth_hh i.mediae_expo i.birth_order i.dma i.V102 i.contextual_regions
> if educ_mom==2 || psu :,or nolog

Mixed-effects logistic regression Number of obs = 315
Group variable: psu Number of groups = 216

Obs per group:
min = 1
avg = 1.5
max = 4

Integration method: mvaghermite Integration pts. = 7

Wald chi2(15) = 21.39
Log likelihood = -90.381795 Prob > chi2 = 0.1249
------------------------------------------------------------------------------------------------
del_place | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------------------------+----------------------------------------------------------------
phy_ipv |
Yes | 4.013633 2.682928 2.08 0.038 1.082801 14.87739
|

The problem with this model is that the confidence interval for the association between physical IPV (phy_ipv) and HFD (health facility delivery place/del_place) is so broad. We can see how broad the odds ratios (AOR = 4.01; CI: 1.08, 14.88). So, I am not sure why there is the broad odds rations?

Thank you so much.