# Identifying Judicial Empathy: Does Having Daughters Cause Judges to Rule for Women’s Issues?

**MEMORANDUM**

**TO: **Justice Clarence Thomas

**FROM: **Mandlenkosi Sibanda

**CC: **Wendy Long *(clerk )*

**DATE: **April 19, 2018

**SUBJECT: **Identifying Judicial Empathy: Does Having Daughters Cause Judges to Rule for Women’s Issues?

**Summary**

This memorandum serves as a follow-up and support to the paper published by Glynn and Sen, (2014) titled *Identifying Judicial Empathy: Does Having Daughters Cause Judges to Rule for Women’s Issues?*. After replicating their analysis and running matching algorithms, I can firmly attest to the validity of their study.

**Randomization**

Glynn and Sen, (2014) argue that their data qualifies the criteria of a natural quasi-experiment because when a family decides to have a child, the sex of the child is outside the parent’s control. This means that their data is free from other individual characteristics like partisanship or ideology (Glynn and Sen, 2014). And because of that, the relationship established between being a judge and having a male or female child is exogenously, and also allows Glynn and Sen, (2014) “to quantify the causal effect of a personal relationship on decision making.” If the units are appropriately randomized as they claim, we would expect to see a similar distribution across all observed and unobserved covariates for both the treatment and control units. We can check this by plotting a histogram for any covariate for both the control and treatment group. For the covariates girls and child, their corresponding graphs are shown below in figure 1.

*Figure 1: Histograms for child and girls covariates, for both treatment and control group*

**Code used**

Figure 1 demonstrates the similarity of the distributions and therefore proves the randomness assumed by Glynn and Sen, (2014).

**Method Replication and Analysis**

Figure 2 shows the graph that Glynn and Sen, (2014) plotted to show that democrat judges voted in a more feminist manner than republican judges.

*Figure 2: Graph showing the proportion of cases decided in a feminist direction. Comparing democrats and republican judges*

I went a step further and plotted the graph of decisions made by judges “with no girls” versus judges “with girls.” Figure 3 shows that judges with daughters voted in a more feminist way than the judges without daughters.

*Figure 3: Graph showing the proportion of cases decided in a feminist direction. Comparing judges with daughters and those without*

**Code used**

**Treatment Effect and Matching**

The treatment is having a daughter and the control group is the group of judges without daughters.

I calculated the treatment effect by subtracting the means of the control group from the treatment group to obtain 0.4733. This means that having a daughter moved the decision of a judge to a more feminist direction by a factor of 47.3%. I then calculated the 95% confidence interval for this estimate using simple linear regression to get 0.356; 0.4576. However, this result had a p-value of 0.187, and since it is higher than 0.05, it is not statistically significant.

**Code used**

To perform matching, I had to change the girls variable to a logical variable of 1s and 0s and then append it to the main data as “girls_A.”

**Code used**

I then used propensity score matching to produce an estimated treatment effect of 0.36 with a 95% confidence of between 0.34 and 0.37. The p-value is also very small (6.596e-12) which makes the result statistically significant. I also ran a multivariate matching procedure that gave a treatment effect of 0.36 and 95% confidence interval of 0.34; 0.38. The p-value was also less than 0.05. I then concluded by running a genetic matching with multivariates which gave a treatment effect of 0.36 and 95% confidence interval of 0.34; 0.38.

**Code used**

**Conclusion**

The treatment effect calculated by using different matching procedures were all positive and had a very small range (confidence interval) and small p-values. It, therefore, shows that the data used and the model ran by Glynn and Sen, (2014) is statistically accurate, thus making their conclusions significant.

Full code available ** here**: