[Weiwen Ng]

Weiwen,

* Example generated by -dataex-. To install: ssc install dataex
clear
input long location str10 date str5 elec_consumption str2 treatgr
600001 "2017-01-01" "66.1" "0"
600003 "2017-01-01" "46.7" "3"
600004 "2017-01-01" "10.8" "3"
600005 "2017-01-01" "24" "3"
600006 "2017-01-01" "42.5" "2"
600007 "2017-01-01" "7.1" "3"
600008 "2017-01-01" "41.1" "1"
600009 "2017-01-01" "41.4" "2"
600010 "2017-01-01" "96.1" "3"
600011 "2017-01-01" "22.1" "0"
600012 "2017-01-01" "31" "1"
600013 "2017-01-01" "33.1" "2"
600014 "2017-01-01" "139" "2"
600015 "2017-01-01" "44.9" "0"
600016 "2017-01-01" "76.9" "2"
600017 "2017-01-01" "34" "0"
600018 "2017-01-01" "4.9" "3"
600019 "2017-01-01" "27.1" "3"
600020 "2017-01-01" "50.5" "1"
600022 "2017-01-01" "47.1" "3"
end

I have data for 2017-2018 (sorted by date). As it has been said, the following R code is supposed to build an event-study figure showing average daily electricity consumption (= average consumption for the relevant treatment group minus the average consumption for the control group) by households in the treated groups (3 treatment groups, variable ‘treatgr’, treatgr=1/2/3) compared to the control group (treatgr=0) over time. The date of the treatment is May 2, 2017 (it should be the vertical line on the figure).

day_avg <- dat %>%
group_by(date, treatgrp) %>%
summarise(mean_consumption = mean(elec_consumption)) %>%
spread(key=treatgrp, value=mean_consumption) %>%
mutate(diff1 = `1` - `0`,
diff2 = `2` - `0`,
diff3 = `3` - `0`) %>%
select(date, contains("diff")) %>%
gather(key=treatment, value=mean_consumption,contains("diff"))

ggplot(day_avg,
aes(x=date, y=mean_consumption, colour=factor(treatment))) +
theme_bw() +
xlab("") +
geom_line() +
ylab("Electricity consumption per day (kWh/day)") +
geom_hline(yintercept=0) +
geom_vline(xintercept = as.numeric(as.Date("02-05-2017","%d-%m-%Y")), linetype="dashed")


I have a follow-up question on this. I guess that this figure will not actually be an event-study one... If I want to draw an event-study figure for my data (i.e., the one which typically displays point estimates from an event study regression of electricity consumption before and after the treatment), how can I do this? I have been desperately trying to find any information about an event study, but all I have found so far is related to financial data.

I would appreciate any help!

I'd post a new thread at this point. In general, you can do this by fitting the regression model and then using margins and marginsplot. However, I don't really know what an event study is. If it's a bit like a difference in differences setup, then what I said should hold.

[Katherine Adams]

Weiwen,

Thank you so much for your help. I read the description of 'margins' and 'marginsplot' - it looks like these commands would be of great help to me. However, I did not get how to use them at all... :-) But I will keep studying.

[Dave Airey]

Note I changed the input example dates to have more than one date, but otherwise this translates the R code, which I'm familiar with.

Dave

* Example generated by -dataex-. To install: ssc install dataex
clear
input long location str10 date str5 elec_consumption str2 treatgr
600001 "2017-01-01" "66.1" "0"
600003 "2017-01-01" "46.7" "3"
600004 "2017-01-01" "10.8" "3"
600005 "2017-01-01" "24" "3"
600006 "2017-01-01" "42.5" "2"
600007 "2017-01-01" "7.1" "3"
600008 "2017-01-01" "41.1" "1"
600009 "2017-01-01" "41.4" "2"
600010 "2017-01-01" "96.1" "3"
600011 "2017-01-01" "22.1" "0"
600012 "2017-03-01" "31" "1"
600013 "2017-03-01" "33.1" "2"
600014 "2017-03-01" "139" "2"
600015 "2017-03-01" "44.9" "0"
600016 "2017-03-01" "76.9" "2"
600017 "2017-03-01" "34" "0"
600018 "2017-03-01" "4.9" "3"
600019 "2017-03-01" "27.1" "3"
600020 "2017-03-01" "50.5" "1"
600022 "2017-03-01" "47.1" "3"
end

rename elec_consumption ec
destring ec, replace force
destring treatgr, replace force
collapse (mean) ec, by(treatgr date)
reshape wide ec, i(date) j(treatgr)
generate diff1 = ec1 - ec0
generate diff2 = ec2 - ec0
generate diff3 = ec3 - ec0
keep date diff*
reshape long diff, i(date) j(comparison)

generate date2 = date(date, "YMD", 2019)
format date2 %td

set scheme s2color
grstyle init
grstyle set imesh
grstyle set color, opacity(50): p#markfill
grstyle set color, opacity(50): p#markline

graph twoway (line diff date2 if comparison == 1) ///
    (line diff date2 if comparison == 2) ///
    (line diff date2 if comparison == 3), ///
    yline(0, lcolor(black%50)) ///
    xline(`=date("2017-02-05", "YMD", 2019)', lcolor(black%50)) ///
    ytitle("Electricity consumption per day (kWh/day)") ///
    legend(label(1 "1-0") label(2 "2-0") label(3 "3-0") ///
    ring(0) position(11) col(1))

[Katherine Adams]

Dave,

Thank you for your answer! I ran the code on my data and got a very interesting figure:



But I will try to figure this out.

[Dave Airey]

Well that is fugly! It looks like in your graph you have just three lines, but each line is going all over the place. How many green lines are you expecting for example? Is the input variable unique per row in your dataset?

[Katherine Adams]

Dave,

Well, the figure they got in R has only one line of each color (so, there are 3 lines in the figure). I am sorry I do not quite understand your last question...

[Dave Airey]

Maybe the data need to be sorted by date and group before plotting?

[Katherine Adams]

Dave,

Yes, I sorted the data by date2, and the result is much better now! Thanks a lot!



I have found almost the same R code but for monthly averages (the code is in red), which might look even better on the graph. I am trying to convert the R commands into Stata using your code, but I am not sure about most of the code lines (my commands are in black). Could you please correct me one more time?



rename lconsum ec
destring ec, replace force
destring treatgr, replace force

month_avg <- dat %>%
mutate(month_of_year = format(date, "%m"),
year = format(date, "%Y")) %>%
group_by(month_of_year,year,treatrp) %>%

summarise(mean_consumption = mean(elec_consumption),
collapse (mean) ec, by(treatgr date)

date = first(date)) %>%
spread(key=treatrp, value=mean_consumption) %>%

mutate(diff1 = `1` - `0`,
diff2 = `2` - `0`,
diff3 = `3` - `0`) %>%
generate diff1 = ec1 - ec0
generate diff2 = ec2 - ec0
generate diff3 = ec3 - ec0

select(date, year, month_of_year, contains("diff")) %>%
gather(key=treatment, value=mean_consumption, contains("diff"))
keep date diff*



Thank you anyway!

[Dave Airey]

I have to say that posting part of the R code mixed into part of the Stata code is not helpful towards understanding. Hopefully you can figure things out.

[Weiwen Ng]

Dave,

Yes, I sorted the data by date2, and the result is much better now! Thanks a lot!
[ATTACH=CONFIG]n1484061[/ATTACH]


I have found almost the same R code but for monthly averages (the code is in red), which might look even better on the graph. I am trying to convert the R commands into Stata using your code, but I am not sure about most of the code lines (my commands are in black). Could you please correct me one more time?



rename lconsum ec
destring ec, replace force
destring treatgr, replace force

month_avg <- dat %>%
mutate(month_of_year = format(date, "%m"),
year = format(date, "%Y")) %>%
group_by(month_of_year,year,treatrp) %>%

summarise(mean_consumption = mean(elec_consumption),
collapse (mean) ec, by(treatgr date)

date = first(date)) %>%
spread(key=treatrp, value=mean_consumption) %>%

mutate(diff1 = `1` - `0`,
diff2 = `2` - `0`,
diff3 = `3` - `0`) %>%
generate diff1 = ec1 - ec0
generate diff2 = ec2 - ec0
generate diff3 = ec3 - ec0

select(date, year, month_of_year, contains("diff")) %>%
gather(key=treatment, value=mean_consumption, contains("diff"))
keep date diff*



Thank you anyway!

Going back to the original csv file, you need to create a monthly date indicator. Then, collapse over month - we both advised you to collapse over date, but I didn't recognize that your R code is in fact summarizing by month. You may have to debug this, but it should work or get you close:

rename elec_consumption ec
destring ec, replace force
destring treatgr, replace force
gen date2 = date(date, "YMD", 2019)
gen month = mofd(date2)

collapse (mean) ec, by(treatgr month)
reshape wide ec, i(month) j(treatgr)
generate diff1 = ec1 - ec0
generate diff2 = ec2 - ec0
generate diff3 = ec3 - ec0
keep month diff*
reshape month diff, i(month) j(comparison)

[Katherine Adams]

Weiwen and Dave,

Thank you for your replies!