CODEWITHSUNDEEP
r
Kashif
Kashif

Reputation: 3327

How to regress on previous time step in R

How can I regress on the previous time step in R here? I want to regress deaths at time t based on deaths, guns, and shootings at time t-1.

City    Year  Month Deaths  Guns     Shootings
Miami   2010    1   69      73800       701        
Miami   2010    2   99      85050       738         
Miami   2010    3   122     92650       784
...
Miami   2013    5   204     99280       800
Miami   2013    6   234     110023      825        
Houston 2011    1   98      92100       789          
Houston 2011    2   146     103900      799         
Houston 2011    3   162     136100      772

Also in R, how do I use ggplot to show the month AND the year on the x-axis? I had a numeric date column in my dataset as well but I couldn't get it to plot by month and year.

Upvotes: 0

Views: 70

Answers (2)

Carl
Carl

Reputation: 7540

You could convert the year and month to a date object using clock's date_build to ggplot the Deaths or fitted values etc. against the date.

And if you wanted separate lm models per City you could nest by City and iterate through the models using purrr's map.

library(tidyverse)
library(clock)
library(broom)

# With some further made-up data
df <- tribble(
      ~City, ~Year, ~Month, ~Deaths,   ~Guns, ~Shootings,
    "Miami", 2010L,     1L,     69,  73800,       701,
    "Miami", 2010L,     2L,     99,  85050,       738,
    "Miami", 2010L,     3L,    122,  92650,       784,
    "Miami", 2010L,     4L,    204,  99280,       800,
    "Miami", 2010L,     5L,    234, 110023,       825,
    "Miami", 2010L,     6L,    244, 110500,       830,
  "Houston", 2011L,     1L,     98,  92100,       789,
  "Houston", 2011L,     2L,    146, 103900,       799,
  "Houston", 2011L,     3L,    162, 136100,       772,
  "Houston", 2011L,     4L,    182, 146100,       782,
  "Houston", 2011L,     5L,    192, 156100,       792
  )

df2 <- df |> 
  group_by(City) |> 
  mutate(date = date_build(Year, Month),
         lead_deaths = lead(Deaths)) |> 
  drop_na() |> 
  nest() |> 
  mutate(model = map(data, ~lm(lead_deaths ~ Deaths + Guns + Shootings, data = .)),
         augmented = map(model, augment),
         fitted = map(augmented, ".fitted")) |> 
  unnest(c(data, fitted)) |> 
  ungroup() 

df2 |> 
  ggplot(aes(date, Deaths)) +
  geom_point() +
  scale_x_date(date_labels = "%Y-%b") +
  facet_wrap(~ City, scales = "free_x") +
  labs(x = NULL) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Created on 2022-07-02 by the reprex package (v2.0.1)

Upvotes: 0

Rui Barradas
Rui Barradas

Reputation: 76432

You can use dplyr::lag to lag the data and then fit the linear model. And, since a dplyr function is being used, I will do it in a pipe.

x <- 'City    Year  Month Deaths  Guns     Shootings
Miami   2010    1   69      73800       701        
Miami   2010    2   99      85050       738         
Miami   2010    3   122     92650       784
Miami   2013    5   204     99280       800
Miami   2013    6   234     110023      825        
Houston 2011    1   98      92100       789          
Houston 2011    2   146     103900      799         
Houston 2011    3   162     136100      772'
df1 <- read.table(textConnection(x), header = TRUE)

suppressPackageStartupMessages(library(dplyr))

fit <- df1 %>%
  mutate(Guns = lag(Guns), Shootings = lag(Shootings)) %>%
  lm(Deaths ~ lag(Deaths) + Guns + Shootings, .)

summary(fit)
#> 
#> Call:
#> lm(formula = Deaths ~ lag(Deaths) + Guns + Shootings, data = .)
#> 
#> Residuals:
#>        2        3        4        5        6        7        8 
#> -17.2273   0.8505  29.1597  53.4768 -63.1397 -38.9155  35.7956 
#> 
#> Coefficients:
#>               Estimate Std. Error t value Pr(>|t|)
#> (Intercept) -9.668e+02  9.909e+02  -0.976    0.401
#> lag(Deaths)  2.567e-01  8.452e-01   0.304    0.781
#> Guns        -8.023e-03  8.976e-03  -0.894    0.437
#> Shootings    2.364e+00  2.152e+00   1.099    0.352
#> 
#> Residual standard error: 59.97 on 3 degrees of freedom
#>   (1 observation deleted due to missingness)
#> Multiple R-squared:  0.3335, Adjusted R-squared:  -0.333 
#> F-statistic: 0.5004 on 3 and 3 DF,  p-value: 0.708

Created on 2022-07-02 by the reprex package (v2.0.1)

Upvotes: 1

Related Questions