Correlation level (r) | HIV+ | HIV− |
|---|
Antenatal | Birth | Postnatal | Antenatal | Birth | Postnatal |
|---|
p value | p value | p value | p value | p value | p value |
|---|
0.05 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 |
0.10 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 |
0.15 | 0.02 | 0.04 | 0.04 | 0.02 | 0.08 | 0.07 |
0.20 | 0.06 | 0.09 | 0.08 | 0.07 | 0.11 | 0.15 |
0.25 | 0.11 | 0.23 | 0.18 | 0.12 | 0.54 | 0.48 |
- Decision rule: significant results (seen by the p value) despite increasing correlation indicate robust results
- To help us see the effect of unobservables on our outcome variable, we focus on the correlation between error terms which capture the extent to which unobservables affect the treatment and the outcome, in our case taking up professional care and eventually the probability of kids surviving beyond age 5. The Altonji test works by assigning different values of correlation (r) and estimating the probit model conditional on these r values. This allows us to see if the parameter changes significantly given the different r values. We allow our r values to range from 0 to 0.25 (by using 0.05 intervals). If the ATT remains significant despite increasing r, then our results are robust
- By checking our p values from table above, we can see that even assuming relatively high selection on unobservables (r = 0.15), the effect of professional care on under-five mortality remains highly statistically significant. This therefore means that sensitivity analysis shows that our results are robust and unlikely to be affected by unobservables
