关键词 > EMET3006/4301/8001
Applied Micro econometrics, EMET3006/4301/8001 Semester 2, 2022 Tutorial 5
发布时间:2022-08-30
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
Applied Micro econometrics, EMET3006/4301/8001
Semester 2, 2022
Tutorial 5 (Week 6)
Write a program that you can use to replicate the figures and tables in Cheng and Hoekstra (2013)
1. Start your script with the regular preamble that installs and loads all the programs you need. See the lecture notes for what you need. (plm package)
2. If you haven’t already then save the castle-doctrine-2000-2010.dta data set on your h-drive, I have a folder called data. (You download the data from Wattle)
3. Peruse the data in R so that you can see what the variables are. Use the commands from Tutorial 0.
4. Table 2 in the paper provides summary statistics for the dependent and control variables. Replicate the unweighted means.
5. Let’s replicate the weighted means. They are weighted by the size of the population. wmean<-summarise(data, weighted .mean(homicide, population), weighted .mean(jhcitizen c, population),
weighted .mean(jhpolice c, population), weighted .mean(robbery, population), weighted .mean(assault, population),
weighted .mean(burglary, population), weighted .mean(larceny, population), weighted .mean(motor, population),
weighted .mean(robbery gun r, population))
Why do we need to weight the means by the size of the population in a state?
6. Log all of the outcome variables for those that are an amount per 100,000 population. You will need to have installed the “logr” program.
lhomicide=log(data$homicide)
ljhcitizen_c =log(data$jhcitizen_c)
ljhpolice_c =log(data$jhpolice_c)
lrobbery =log(data$robbery)
lassault =log(data$assault)
lburglary =log(data$burglary)
llarceny =log(data$larceny)
lmotor =log(data$motor)
lrobbery_gun_r=log(data$robbery_gun_r)
lpolice=log(police)
lprisoner=log(prisoner)
llagprisoner=log(lagprisoner)
7. Replicate the first column in the unweighted part of Table 3 for larceny. What do you find? Use either pooled OLS with fixed effects and clus- tered standard errors or convert the data to panel data. My lecture slides have this code.
8. Continue adding variables as in the specification in Table 3. Discuss the coefficient on cdl in each specification. (I can’t get the exact same results for some of the columns, don’t let this worry you if it happens to you.)
9. Rerun everything using population as the weight. Discuss the coeffi- cient on cdl.
#With weights
pooling_weights <- plm(llarceny˜cdl+
factor(sid) + factor(year), data=pdata,
weights = population, model= "pooling") summary(pooling_weights)
10. Run the regressions in Tables 4 and 5. (Do some of this so you get the hang of it, don’t spend ages doing it)
11. Discuss the results of the regression in terms of placebo, deterrence and homicides.
12. Replicate the graphs in figure 1 and discuss. (I regret asking this.) Here is a code I came up with that just does 2005 for figure 1. Ideally I would have points joined by lines but I gave up. Figure 2 is optional.
#Figure 1
# I’ll just do one graph from figure 1
# I’m using the Castle data which I’ve renamed data and I’ve logged the variables from earlier, everything else is the same as in the original data set .
#We’ll use the effective year variable for this but some # states never change their law and so their value is NA # We want it to be zero instead .
data[is .na(data)] = 0
data<-data %>%
mutate(year_2005=ifelse(effyear==2005, 1, 0),
control=ifelse(effyear==0, 1, 0))
data_reduce<-data[c("year", "sid", "lhomicide", "effyear", "control", "year_2005")]
data_reduce<-data_reduce %>%
group_by(year, control) %>%
mutate(mean_lhom_control=mean(lhomicide))
data_reduce<-data_reduce %>%
group_by(year,year_2005) %>%
mutate(mean_lhom_2005=mean(lhomicide))
#I’m making a new dataset of just the means . mean_lhom_control<-aggregate(data_reduce$lhomicide, by=list(data_reduce$year, data_reduce$control), FUN=mean) mean_lhom_2005<-aggregate(data_reduce$lhomicide, by=list(data_reduce$year, data_reduce$year_2005), FUN=mean) mean_lhom_control<-subset(mean_lhom_control, Group .2>0) mean_lhom_2005<-subset(mean_lhom_2005, Group .2>0)
mean_2005<-merge(mean_lhom_control, mean_lhom_2005, by .x="Group . 1", by .y="Group . 1")
ggplot(aes(Group . 1), data=mean_2005) +
geom_line(aes(y=x .x), colour="red")+
geom_line(aes(y=x .y), colour="blue")+
geom_vline(xintercept = 2005, colour = "black", linetype = 2)+ xlab("Year") +
ylab("log(Homicide Rate)") + ylim(1, 2)
13. What effect does this analysis say reforming castle doctrine laws has on homicides?
14. What are the key parts of these legislative reforms that you think may be causing this result?
15. Explain what SUTVA requires in order for these estimates to be causal?
16. Assume there are spillovers to neighboring states created by castle doc- trine reforms. Does that imply that Cheng and Hoekstra’s result is too large or too small? Why/why not?