It has been commonly argued in sociology that odds ratio—as a measure of association between categorical variables—is appealing, because of its “margin-free” property. This has always baffled me, so I decided to articulate my bafflement here as a sanity check.

Inverse probability weighting (IPW) is a popular tool for estimating the average treatment effect (ATE) of a binary variable under the conditional ignorability assumption. There are multiple variants of IPW. Particularly, I used to be intrigued by the relationship between the so-called Horvitz–Thompson (HT) estimator and the Weighted Least Squares (WLS) estimator, both of which implement IPW. Both estimators are introduced in popular causal inference textbooks. For example, Angrist and Pischke (2009, p.82) focus on the HT estimator, while Winship and Morgan (2014, p228-9) focus on the WLS estimator. However, I haven’t seen them introduced together in these textbooks, and their relationship could seem a little unclear. In fact, a quick simulation would reveal that the classic version of the HT estimator gives a different ATE estimate from that of the WLS estimator in any finite sample (although asymptotically they both converge to the true ATE).

It has been commonly argued in sociology that odds ratio—as a measure of association between categorical variables—is appealing, because of its “margin-free” property. This has always baffled me, so I decided to articulate my bafflement here as a sanity check.

Inverse probability weighting (IPW) is a popular tool for estimating the average treatment effect (ATE) of a binary variable under the conditional ignorability assumption. There are multiple variants of IPW. Particularly, I used to be intrigued by the relationship between the so-called Horvitz–Thompson (HT) estimator and the Weighted Least Squares (WLS) estimator, both of which implement IPW. Both estimators are introduced in popular causal inference textbooks. For example, Angrist and Pischke (2009, p.82) focus on the HT estimator, while Winship and Morgan (2014, p228-9) focus on the WLS estimator. However, I haven’t seen them introduced together in these textbooks, and their relationship could seem a little unclear. In fact, a quick simulation would reveal that the classic version of the HT estimator gives a different ATE estimate from that of the WLS estimator in any finite sample (although asymptotically they both converge to the true ATE).

It has been commonly argued in sociology that odds ratio—as a measure of association between categorical variables—is appealing, because of its “margin-free” property. This has always baffled me, so I decided to articulate my bafflement here as a sanity check.

Inverse probability weighting (IPW) is a popular tool for estimating the average treatment effect (ATE) of a binary variable under the conditional ignorability assumption. There are multiple variants of IPW. Particularly, I used to be intrigued by the relationship between the so-called Horvitz–Thompson (HT) estimator and the Weighted Least Squares (WLS) estimator, both of which implement IPW. Both estimators are introduced in popular causal inference textbooks. For example, Angrist and Pischke (2009, p.82) focus on the HT estimator, while Winship and Morgan (2014, p228-9) focus on the WLS estimator. However, I haven’t seen them introduced together in these textbooks, and their relationship could seem a little unclear. In fact, a quick simulation would reveal that the classic version of the HT estimator gives a different ATE estimate from that of the WLS estimator in any finite sample (although asymptotically they both converge to the true ATE).