tests/testthat/test-ice.R

In audreyrenson/paralleltrends: Estimators for sustained treatment effects under parallel trends assumptions

test_that("ice wide and long return same result", {
  Tt = 3
  Beta = generate_parameters(Tt=Tt, range_ymeans=stats::qlogis(c(0.01, 0.05)))
  N = 100

  df_wide = generate_data(N, Tt, Beta)
  df_wide_interv = df_wide %>% dplyr::mutate(dplyr::across(dplyr::starts_with('A'), ~ 0))

  ice_wide = ice(df_obs=df_wide,
                 df_interv = df_wide_interv,
                 inside_formula_t = '~A{t}*(W1{t}+W2{t}+I(W2{t}^2))',
                 inside_formula_tmin1 = '~A{t}*(W1{t-1}+W2{t-1}+I(W2{t-1}^2))',
                 outside_formula = '~A{k}*(W1{k}+W2{k}+I(W2{k}^2))',
                 Tt=Tt,
                 models=FALSE,
                 suppress_rank_warnings=TRUE,
                 long=FALSE)



  df_long = pivot_longer_gendata(df_wide) %>%
    dplyr::mutate(W22=W2^2) %>%
    dplyr::group_by(uid) %>%
    append_lags(3, c('A','W1','W2','W22', 'Y'), default=NA) %>%
    dplyr::ungroup() %>%
    make_time_dummies('t')

  ice_long = ice(df_long,
                 df_interv = df_long %>% dplyr::mutate(dplyr::across(dplyr::starts_with('A'), ~0)),
                 inside_formula_t = 'Y~-1 + {tvars}*(W1+W2+W22)*A - A*(W1+W2+W22)',
                 inside_formula_tmin1 ='Y_lag1~ -1 + {tvars}*(W1_lag1+W2_lag1+W22_lag1)*A - (W1_lag1+W2_lag1+W22_lag1)*A',
                 outside_formula = 'preds ~ -1 + {tvars}*(W1_lag{n}+W2_lag{n}+W22_lag{n})*A_lag{n} -(W1_lag{n}+W2_lag{n}+W22_lag{n})*A_lag{n} - t{n}:A_lag{n} - t{n}:(W1_lag{n}+W2_lag{n}+W22_lag{n}):A_lag{n} ',
                 inside_family=gaussian,
                 t_col=df_long$t,
                 Tt=Tt,
                 models=FALSE,
                 suppress_rank_warnings=TRUE,
                 long=TRUE)


  expect_equal(ice_wide, ice_long)
})