Produce a meaningful error message when the outcome has no variation

[sou412]

When the control group has no variation in the outcome during any of the pre-treatment period (in the range of lag value), did_check() function fails with the following error message"

 # Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) : 
 #   NA/NaN/Inf in 'y'

This is because of the outcome normalization happening here: https://github.com/naoki-egami/DIDdesign/blob/8440250a4259fd3e8b5b616cf87f2333491c4ca2/R/did-check-std.R#L130

The immediate fix is to produce a more meaningful error message when this happens.

[sou412]

Test code before fix.

library(dplyr)
library(testthat)
data(anzia2012)
set.seed(1234)

# create a test data ----------------------------------------------------------
treated_id <- anzia2012 %>%
  filter(year == 2007, oncycle == 1) %>%
  pull(district)
test_data <- anzia2012 %>%
  filter(year %in% c(2005, 2006, 2007)) %>%
  mutate(
    outcome_var0_control = lnavgsalary_cpi,
    outcome_var0_treatment = lnavgsalary_cpi,
    outcome_var0_all = lnavgsalary_cpi
  ) %>%
  bind_rows(
    anzia2012 %>%
      filter(year == 2004) %>%
      mutate(
        outcome_var0_control = case_when(
          district %in% treated_id ~ lnavgsalary_cpi,
          TRUE ~ 0
        ),
        outcome_var0_treatment = case_when(
          !(district %in% treated_id) ~ lnavgsalary_cpi,
          TRUE ~ 0
        ),
        outcome_var0_all = 0
      )
  )

# run the function ------------------------------------------------------------
## Expected to fail because var(Y|D=0) = 0 for t-2
expect_error(
  did_check(
    formula = outcome_var0_control ~ oncycle,
    data    = test_data,
    id_unit = "district",
    id_time = "year",
    option  = list(n_boot = 20, parallel = TRUE, lag = 1:2)
  ),
  "NA/NaN/Inf in 'y'"
)

## Expected to fail because var(Y|D=0) = 0 for t-2
expect_error(
  did_check(
    formula = outcome_var0_all ~ oncycle | teachers_avg_yrs_exper +
      ami_pc + asian_pc + black_pc + hisp_pc,
    data = test_data,
    id_unit = "district",
    id_time = "year",
    option = list(n_boot = 20, parallel = TRUE, lag = 1:2)
  ),
  "NA/NaN/Inf in 'y'"
)

## Expected to run because var(Y|D=0) != 0 for t-1
expect_type(
  did_check(
    formula = outcome_var0_control ~ oncycle,
    data    = test_data,
    id_unit = "district",
    id_time = "year",
    option  = list(n_boot = 20, parallel = TRUE, lag = 1)
  ),
  "list"
)

## Expected to run
expect_type(
  did_check(
    formula = outcome_var0_treatment ~ oncycle | teachers_avg_yrs_exper +
      ami_pc + asian_pc + black_pc + hisp_pc,
    data = test_data,
    id_unit = "district",
    id_time = "year",
    option = list(n_boot = 20, parallel = TRUE, lag = 1:2)
  ),
  "list"
)

expect_type(
  out_var0_treatment <- did_check(
    formula = outcome_var0_treatment ~ oncycle | teachers_avg_yrs_exper +
      ami_pc + asian_pc + black_pc + hisp_pc,
    data = test_data,
    id_unit = "district",
    id_time = "year",
    option = list(n_boot = 20, parallel = TRUE, lag = 1:2)
  ),
  "list"
)

summary(out_var0_treatment)

When it runs, it returns estimates:

> summary(out_var0_treatment)
── Estimates for assessing parallel trends assumption ─────────────────────────────────────────────────────────────────────────────────────────
  estimate lag std.error EqCI95_LB EqCI95_UB
1 -0.00361   1   0.00178   -0.0847    0.0847
2 10.63308   2   0.00611 -145.3014  145.3014