In the picture, I've attached below, the gamma value of 1.2 (Row 3) shows the result of Rosenbaum sensitivity analysis based on Wilcoxon sign rank test (Rosenbaum 2002).
If I understand this right, an intuitive interpretation of this statistic is that the outcome estimate would remain significantly different from zero (p < 0.05) in the presence of a hidden bias that could cause the odds of being affected by treatment assignment, to differ by a factor as high as 1.2. In other words, even if the odds of one household receiving treatment/ intervention is only 1.2 times higher because of different values on unobserved covariates despite being identical on the matched covariates, our inference changes.
This result also indicates that 90% confidence interval of the HL estimate would still exclude zero in the presence of an additive fixed hidden bias that could cause the odds of being affected to differ by a factor of 1.3.
The general conclusion then is that while it would appear that the intervention has a positive treatment effect, the finding is sensitive to possible hidden bias due to an unobserved confounder. Am I right?
If I understand this right, an intuitive interpretation of this statistic is that the outcome estimate would remain significantly different from zero (p < 0.05) in the presence of a hidden bias that could cause the odds of being affected by treatment assignment, to differ by a factor as high as 1.2. In other words, even if the odds of one household receiving treatment/ intervention is only 1.2 times higher because of different values on unobserved covariates despite being identical on the matched covariates, our inference changes.
This result also indicates that 90% confidence interval of the HL estimate would still exclude zero in the presence of an additive fixed hidden bias that could cause the odds of being affected to differ by a factor of 1.3.
The general conclusion then is that while it would appear that the intervention has a positive treatment effect, the finding is sensitive to possible hidden bias due to an unobserved confounder. Am I right?
