Main linear regressions

Note

In this file I run the main regressions with the different measures of temperature and reports data.

• I export “Table 1”
• I save the plots with the estimates for the quintiles of precipitation, humidity and windspeed.

start_time <- Sys.time()

library(tidyverse)
library(here) # path control
library(fixest) # estimations with fixed effects
library(knitr) 
library(patchwork)
library(tictoc)
tic("Total render time")

data <-
  read_rds(here("..", "output", "data_reports.rds")) %>% 
  filter(covid_phase == "Pre-pandemic")

# quintiles are computed last minute, on the selected period
data <- data %>% 
  mutate(prec_quintile = ntile(prec, 5),
         rh_quintile = ntile(rh, 5),
         wsp_quintile = ntile(wsp, 5))

data <- data %>% 
  left_join(
    read_rds(here("..", "output", "temporary_data",
                  "tmp_hourly_ageb.rds"))
  )

Run the regressions

Tmean

First, run the baseline regression with tmean, 24-hour mean of temperature

reg_linear_reports <- data %>% 
  fepois(reports_dv ~ tmean |
           prec_quintile + rh_quintile + wsp_quintile +
           ageb + year^month + day_of_week + day_of_year,
         cluster = ~ ageb)

Tmin and Tmax

Use tmin and tmax instead

reg_min <- data %>% 
  fepois(reports_dv ~ tmin |
           prec_quintile + rh_quintile + wsp_quintile +
           ageb + year^month + day_of_week + day_of_year,
         cluster = ~ ageb)

reg_max <- data %>% 
  fepois(reports_dv ~ tmax |
           prec_quintile + rh_quintile + wsp_quintile +
           ageb + year^month + day_of_week + day_of_year,
         cluster = ~ ageb)

Or both in the same reg:

reg_min_max <- data %>% 
  fepois(reports_dv ~ tmin + tmax |
           prec_quintile + rh_quintile + wsp_quintile +
           ageb + year^month + day_of_week + day_of_year,
         cluster = ~ ageb)

Day and night temperatures

Max temperature is usually reached at 4-5pm

# max hour is always during the daytime
data %>% 
  filter(!is.na(tmax)) %>% 
  select(ageb, date, tmp_hourly) %>%
  # indicator for which hour is the min in a day
  mutate(max_hour = map_dbl(tmp_hourly, ~ which.max(.x))) %>% 
  ggplot() +
  geom_histogram(aes(x = max_hour), bins = 24)

Min temperature is usually reached at hour 7, i.e from 06:00 to 06:59.

# hour in which the temp reaches its min is sometimes during daytime (hour 8, from 07:00 to 07:59)
data %>% 
  filter(!is.na(tmin)) %>% 
  select(ageb, date, tmp_hourly) %>%
  # indicator for which hour is the min in a day
  mutate(min_hour = map_dbl(tmp_hourly, ~ which.min(.x))) %>% 
  ggplot() +
  geom_histogram(aes(x = min_hour), bins = 24)

Since we have hourly measurements of temperature, we can calculate the average temperature during the day and during the night.

I define

• day temperature from 07:00 (hour 8) to 19:59 (hour 20)
• night temperature from 20:00 (hour 21) (the day before!) to 06:59 (hour 7)

So, there are 13 hours in a day, and 11 hours in the night

# I want to account for a different definition of night.
# now if i want the night to be LAGGED NIGHT TEMPERATURE:
# For a crime on Monday, night temperatures reflect Sunday 8:00 PM to Monday 6:59 AM.
data <- data %>% 
  mutate(tmean_day = map_dbl(tmp_hourly, ~ mean(.x[8:20])),
         #this would be the "wrong i.e. not lagged" definition of night
         #tmean_night = map_dbl(tmp_hourly, ~ mean(c(.x[21:24], .x[1:7]))),
         tmean_lagged_night = map2_dbl(
           lag(tmp_hourly, default = list(rep(NA, 24))),  # Provide a default list of NAs
           tmp_hourly,
           ~ if (!all(is.na(.x))) mean(c(.x[21:24], .y[1:7])) else NA_real_)
  )

Here’s the distribution of day and night temperatures.

data %>% 
  ggplot() +
  geom_density(aes(x = tmean_day), color = "orange") +
  geom_density(aes(x = tmean_lagged_night), color = "darkblue") +
  theme_light()

Now i can run the regs, separately for day and night temperatures, and jointly.

reg_day <- data %>% 
  fepois(reports_dv ~ tmean_day |
          prec_quintile + rh_quintile + wsp_quintile +
          ageb + year^month + day_of_week + day_of_year,
        cluster = ~ ageb)

reg_night <- data %>% 
  fepois(reports_dv ~ tmean_lagged_night |
          prec_quintile + rh_quintile + wsp_quintile +
          ageb + year^month + day_of_week + day_of_year,
        cluster = ~ ageb)

reg_day_night <- data %>% 
  fepois(reports_dv ~ tmean_day + tmean_lagged_night |
          prec_quintile + rh_quintile + wsp_quintile +
          ageb + year^month + day_of_week + day_of_year,
        cluster = ~ ageb)

Midnight temperature

Just to try, but doesnt add much

reg_midnight <- data %>% 
  mutate(midnight_temp = map_dbl(tmp_hourly, ~ .x[1])) %>% 
  fepois(reports_dv ~ midnight_temp |
           prec_quintile + rh_quintile + wsp_quintile +
           ageb + year^month + day_of_week + day_of_year,
         cluster = ~ ageb)

Export Table 1

These are the results we want to export:

etable(reg_linear_reports,
       reg_day, reg_night, reg_day_night,
       reg_max, reg_min, reg_min_max) %>% 
  kable()

	reg_linear_reports	reg_day	reg_night	reg_day_night	reg_max	reg_min	reg_min_max
Dependent Var.:	reports_dv	reports_dv	reports_dv	reports_dv	reports_dv	reports_dv	reports_dv
tmean	0.0274*** (0.0034)
tmean_day		0.0246*** (0.0030)		0.0220*** (0.0041)
tmean_lagged_night			0.0188*** (0.0031)	0.0039 (0.0042)
tmax					0.0198*** (0.0026)		0.0159*** (0.0030)
tmin						0.0178*** (0.0029)	0.0097** (0.0034)
Fixed-Effects:	——————	——————	——————	——————	——————	——————	——————
prec_quintile	Yes	Yes	Yes	Yes	Yes	Yes	Yes
rh_quintile	Yes	Yes	Yes	Yes	Yes	Yes	Yes
wsp_quintile	Yes	Yes	Yes	Yes	Yes	Yes	Yes
ageb	Yes	Yes	Yes	Yes	Yes	Yes	Yes
year-month	Yes	Yes	Yes	Yes	Yes	Yes	Yes
day_of_week	Yes	Yes	Yes	Yes	Yes	Yes	Yes
day_of_year	Yes	Yes	Yes	Yes	Yes	Yes	Yes
__________________	__________________	__________________	__________________	__________________	__________________	__________________	__________________
S.E.: Clustered	by: ageb	by: ageb	by: ageb	by: ageb	by: ageb	by: ageb	by: ageb
Observations	3,624,826	3,624,815	3,624,809	3,624,797	3,624,826	3,624,826	3,624,826
Squared Cor.	0.01333	0.01333	0.01332	0.01333	0.01333	0.01332	0.01333
Pseudo R2	0.05612	0.05612	0.05609	0.05612	0.05611	0.05609	0.05612
BIC	804,221.2	804,221.6	804,250.0	804,235.2	804,231.2	804,252.7	804,237.2

Export a latex with options. I have to reformat that table manually.

source(here("fixest_etable_options.R"))

etable(reg_linear_reports,
       reg_day, reg_night, reg_day_night,
       reg_max, reg_min, reg_min_max,
       file = here("..", "output", "tables", "reg_diff_temperature_measures.tex"), replace = T,
       adjustbox = TRUE, float = T,
       title = "The Effect of Temperature on Reported DV",
       label = "reg_diff_temperature_measures",
       notes = "Standard errors clustered by neighborhood in parentheses. * denotes significance at the 10% level, ** at the 5% level, and *** at the 1% level. Panel A presents estimates from the baseline specification (Equation~\ref{eq_main_spec}), where temperature is measured as the mean of 24 hourly readings. Panel B replaces mean temperature with daytime (7:00–20:00) and nighttime (21:00–7:00) temperatures. Column (i) includes only daytime temperature; Column (ii) includes only nighttime temperature; Column (iii) includes both jointly. Panel C adopts the same approach using the minimum and maximum of the 24 hourly readings instead. The dependent variable is the count of reported domestic violence incidents per neighborhood per day."
)

Plots for the quintiles of prec, hum, wsp

Now, we want to see the estimates for the quintiles. A table would be too cumbersome, so I do some plots:

reg_for_the_quintiles <- data %>% 
  fepois(reports_dv ~ tmean +
           i(prec_quintile) + i(rh_quintile) + i(wsp_quintile) |
           ageb + year^month + day_of_week + day_of_year,
         cluster = ~ ageb)

prec_plot <- 
  iplot(reg_for_the_quintiles, only.params = T, i.select = 1)$prms %>% 
  ggplot(aes(x = estimate_names, y = estimate, ymin = ci_low, ymax = ci_high, group = 1)) +
  geom_point(color = "black", size = 2) + geom_errorbar(color = "black", width = .2) +
  geom_hline(yintercept = 0, alpha = .5) +
  theme_light() +
  labs(x = "Precipitation quintile", y = "Estimate and 95% CI")

hum_plot <- 
  iplot(reg_for_the_quintiles, only.params = T, i.select = 2)$prms %>% 
  ggplot(aes(x = estimate_names, y = estimate, ymin = ci_low, ymax = ci_high, group = 1)) +
  geom_point(color = "black", size = 2) + geom_errorbar(color = "black", width = .2) +
  geom_hline(yintercept = 0, alpha = .5) +
  theme_light() +
  labs(x = "Humidity quintile", y = NULL)

wsp_plot <-
  iplot(reg_for_the_quintiles, only.params = T, i.select = 3)$prms %>% 
  ggplot(aes(x = estimate_names, y = estimate, ymin = ci_low, ymax = ci_high, group = 1)) +
  geom_point(color = "black", size = 2) + geom_errorbar(color = "black", width = .2) +
  geom_hline(yintercept = 0, alpha = .5) +
  theme_light() +
  labs(x = "Windspeed quintile", y = NULL)

prec_plot + hum_plot + wsp_plot

ggsave(path = here("..", "output", "figures"),
       filename = "reg_quintiles.png",
       width = 7, height = 2.3)

Rendered on: 2025-07-27 01:21:38

Total render time: 401.474 sec elapsed