Question 1

The following is a sketch of a potential answer. The best responses included the statistics and causal estimates shown in class related to the witch-killing paper as well as supporting evidence from outside sources.

What role do you think local insecurity and violence could be playing?

●  The article suggests that the rise of witch-killings comes from weak governance.  Mob justice was outlawed in 2014, but there has been no enforcement of the law, most likely because of ongoing conflict between armed groups in the region.

●  Perhaps the so-called prophetesses target  “witches” so that if violence reached their village,  the prophetesses would have a reason to be spared, as they can contribute to their society.

What role do you think local economic conditions could be playing?

●  As we saw in lecture, the victims in Tanzania came from relatively poor households, and as elderly women, were the least productive members in that society.  During the lean season, there may be a pressure to eliminate the least productive and then have more resources to share among the rest of the village.

●  This could be a motivating factor in this region of the DRC, especially if the up-tick in witch accusations aligned with extreme weather events or pre-harvest seasons.

Make sure to discuss and possibly contrast the situation described in the article to the issue of witch-killing in rural Tanzania presented in lecture.

● In both settings, elderly women were targeted and accused of being witches. In the case of Tanzanzia that we saw in lecture, the women were blamed for bad rain and crop failure. From what we see in the Guardian article, those “detecting" the witches are doing so for reputation and to grow a following.

Question 2

2a

N/A

2b

#preparing data

# this reads the data

ps3  <- read_csv("Econ172_S22_PS3_data.csv")

# this creates the total murders variable

ps3$total_murders  <- ps3$witch_murders + ps3$oth_murders

#sum(ps3$total_murders==0)/length(ps3$total_murders)

stargazer(as.data.frame (ps3), mean.sd=TRUE  , min.max  =  TRUE  , median  =  TRUE  ,  type  ="latex")

% Table created by stargazer v.5.2.2 by Marek Hlavac, Harvard University. E-mail: hlavac at fas.harvard.edu % Date and time: Thu, Apr 28, 2022 - 02:44:29 PM

Table 1:

Table 1 shows the summary statistics of our data set. Coincidentally, non-witch and witch murders have the same average in our sample.  We can see, however, that both variables are rare events as observed in the mean of 0.09 murders for any given year-village pair. In other words, less than one murder on average was observed in any given village of the sample during our period of study (0.091*11 years ≈ 1).

The any_rain and any_disease variables are also rare, with a mean of 17 and 15% chance of likelihood for any year in any village. Although the rare occurrence of these variables is not an issue per se, it is helpful to keep in mind that these events are rare, as a lack of variation in both outcome and explanatory variable might be problematic for statistical power in the proceeding analysis.

reg1  <- felm(total_murders ~ any_rain|0 |0 |vid , data=ps3)

reg2  <- felm(total_murders ~ any_rain + educat + trad_relig |0 |0 |vid , data=ps3)

reg3  <- felm(total_murders ~ any_disease |0 |0 |vid , data=ps3)

reg4  <- felm(total_murders ~ any_disease + educat + trad_relig |0 |0 |vid , data=ps3)

stargazer(reg1  ,reg2,reg3,reg4,  out= "Table  1" ,type="latex" ,header=TRUE ,  table.placement  =  "!h" ,  title= "

% Table created by stargazer v.5.2.2 by Marek Hlavac, Harvard University. E-mail: hlavac at fas.harvard.edu % Date and time: Thu, Apr 28, 2022 - 02:44:29 PM % Requires LaTeX packages: dcolumn

2c

Columns 1 and 2 of Table 2 present our estimates. We see from the coefficient of any_rain that the occurrence of flood or drought in a given village has a small impact on total murders (β=0.048). The literal interpretation of this coefficient would be that the event of a flood or drought would lead to an increase of 0.048 average murders. Additionally, it is also not statistically significant (t=0.048/0.046 < 1.96). We should interpret this as the fact that, within the context of our experiment, extreme climate events in year t are not correlated with number of murders in year t at the village level.

Adding controls such as education and share of households practicing traditional religion, reduces the effect of any_rain (β=0.04), with the coefficient still statistically insignificant (t=0.04/0.043 < 1.96). Every additional year in the average education of the village decreases total murders by 0.038, but is not statistically significant (-0.038/0.026 < 1.96). Traditional religion is positive and of almost negligent magnitude (0.001), and also not statistically significant (0.001/0.104 < 1.96).