Skip to contents

Supported Models

Source: vignettes/compatability.Rmd
compatability.Rmd

This vignette shows model engines that DAGassist has been verified to work with and provides concrete usage examples. The list is illustrative, not exhaustive. DAGassist is model-agnostic: if an engine accepts a standard formula + data interface, it will usually work.

If you want support documented (or added) for an engine not shown here, please open an issue or PR on GitHub.

Package Function
estimatr lm_robust
fixest feglm
fixest feols
lfe felm
lme4 glmer
lmerTest lmer
MASS glm.nb
stats lm

Notes

  • Clusters: You may pass cluster variables as a bare name, a string, or a one-sided formula (e.g., clusters = cat_b, "cat_b", or ~cat_b). DAGassist resolves them within data.
  • Fixed/Random effects: Random-effects bars (e.g., (1 | group)) and fixest FE/IV tails (| fe ...) in your formula are preserved.
  • Difference-in-Differences: DiD notation works, but the interaction term must be included as a node in the DAG, named exactly as in the model, and set as the exposure. DAGassist breaks if the exposure is missing or mismatched.

Example usage

estimatr::lm_robust 

DAGassist(
    dag = dag_model,
    formula = estimatr::lm_robust(
        ancvs_change_1116 ~ western_belt +
            unemploy_broad + black_popshare + col_popshare +
            indasian_popshare + other_popshare + share_formal +
            share_informal + share_traditional + afrikaans_popshare +
            english_popshare + isindebele_popshare + isixhosa_popshare +
            sepedi_popshare + sesotho_popshare + setswana_popshare +
            signlanguage_popshare + siswati_popshare + tshivenda_popshare +
            xitsonga_popshare +
            latitude + longitude + latitudelongitude,
        data     = df,
        clusters = cat_b)
)

fixest::feglm 

DAGassist(
    dag = dag_model2,
    formula = fixest::feglm(
        exc_eq ~ cat4_prevalence_best + nonvprotgovt + v2x_gencs +
            WINGOImputed_fill + UNintervention + majpowintervention + UNwomen + 
            religious + ideolleft + lgt + substantive + v2x_gencl + 
            Krause_wpm_neg + eqgen_binary,
        data = df %>% filter(interstateonly==0, inter_intrastate==0),
        family = binomial("logit"),
        cluster = ~cowcode
    )
)

fixest::feols 


DAGassist(
  dag = dag_model,
  formula = fixest::feols(
    psoe ~ time * treated2 + log_pop + no_education + growthunemployment+ 
      men_50_rs + immigration_rs + ebal_w | province,
    data = df,
    cluster = ~municipality)
)

lfe::felm 

DAGassist(dag=dag_model,
          formula = lfe::felm(
              share_female ~ lag_democracy_stock_poly_95 + lag_v2xcl_rol + 
                  lag_v2xcl_prpty + lag_v2x_rule + lag_v2x_jucon +
                  lag_v2xlg_legcon + lag_v2x_corr + lag_v2clstown + 
                  lag_v2xcs_ccsi + lag_v2xps_party | year + country_isocode | 0 | 
                  country_isocode,
              data=df_cross)
)

lme4::glmer 


DAGassist(
    dag = dag_model,
    formula = lme4::glmer(
        dsu_file ~ specific + lnearnings + distortion + progress + duration +
            eu + japan + mexico + korea + nonoecd + 
            lntotdirx + lnprod + lnpolcon + active301 +
            (1 | isic3_4dig),
        data   = df,
        family = binomial(link = "logit"),
        control = glmerControl(optimizer = "bobyqa"))
)

lmerTest::lmer 


DAGassist(
    dag = dag_model,
    formula = lmerTest::lmer(
        ekloges_PERCENTAGE_May_2012_GD ~ ekloges_PERCENTAGE_2009_GD +
            N_AntifaEvent_notzero2_Oc09_May12 + avg_age + log_population  +
            prop_ksenoi + (1|code_perif), 
        data=dta))

MASS::glm.nb 

DAGassist(
  dag = dag_model,
  formula = MASS::glm.nb(
    protestcount ~ civilcas_percapita + sigact_percapita + log_pop + shiapop_perc
      kurdpop_perc + urban + totaloilvolx + lognearestcity + feb2011 + nov2010 +
      laggedprotestcount + illiterate + incomequintilelowest + 
      incomequintilehighest + urate + powercontinuous,
    data = iraqevents_timeseries_small,
    control = glm.control(maxit = 50),
    link = "log")
)

stats::lm 

DAGassist(
  dag = dag_model,
  formula = stats::lm(
    restraint ~ nonvio2 + prime_frat + prime_future + prime_RU + prime_UN + 
      oppose92coup2 + civwar + islamist + supp_sharia + active + conscript + 
      soldier + junoff + senoff + branch2 + train_west + train_russia + 
      train_china + young + sex + edu + urban_area + employed + student + 
      arabic + econ1 + corr1 + dem + supp_opp_parties + continue + protested + 
      preboutef + post_exp + mon + factor(fgov), 
    data=data)
)