Construction des
variables (traduit du do-file Stata)
Variables
structurelles
# Variables de traitement et de strate
el <- el %>%
mutate(
pair_id = as.factor(paire),
treated = as.integer(treatment),
# Identifiant village : on utilise paire + treatment comme proxy
village_id = as.factor(paste0(paire, "_", treatment))
)
Accès au crédit
(Section I du questionnaire)
# Le do-file utilise i3_j (type de prêteur) et i9_j (montant)
# pour j = 1 à 5 (prêts actifs 1-3, passés 4-5)
# Codes : 3=Al Amana, 1/2/15=autre formel, 7-14=informel
# Indicatrices de prêt : Al Amana, autre formel, informel
for (j in 1:3) {
el[[paste0("alamana_a", j)]] <- as.integer(el[[paste0("i3_", j)]] == 3)
el[[paste0("other_formal_a", j)]] <- as.integer(el[[paste0("i3_", j)]] %in% c(1, 2, 15))
el[[paste0("informal_a", j)]] <- as.integer(el[[paste0("i3_", j)]] %in% 7:14)
el[[paste0("oamc_a", j)]] <- as.integer(el[[paste0("i3_", j)]] %in% c(4, 5, 6))
}
for (j in 4:5) {
el[[paste0("alamana_p", j)]] <- as.integer(el[[paste0("i3_", j)]] == 3)
el[[paste0("other_formal_p", j)]] <- as.integer(el[[paste0("i3_", j)]] %in% c(1, 2, 15))
el[[paste0("informal_p", j)]] <- as.integer(el[[paste0("i3_", j)]] %in% 7:14)
el[[paste0("oamc_p", j)]] <- as.integer(el[[paste0("i3_", j)]] %in% c(4, 5, 6))
}
# Nombre total de prêts (actifs + passés)
el <- el %>% mutate(
loans_alamana = rowSums(select(., starts_with("alamana_")), na.rm=TRUE),
loans_oformal = rowSums(select(., starts_with("other_formal_")), na.rm=TRUE),
loans_informal = rowSums(select(., starts_with("informal_")), na.rm=TRUE),
loans_oamc = rowSums(select(., starts_with("oamc_")), na.rm=TRUE)
)
# Dummies : a-t-on emprunté au moins une fois ?
el <- el %>% mutate(
borrowed_alamana = as.integer(loans_alamana >= 1),
borrowed_oformal = as.integer(loans_oformal >= 1),
borrowed_informal = as.integer(loans_informal >= 1),
borrowed_total = as.integer((loans_alamana + loans_oformal +
loans_informal + loans_oamc) >= 1)
)
# Montants des prêts actifs (j=1 à 3)
for (j in 1:3) {
i3 <- el[[paste0("i3_", j)]]
i9 <- el[[paste0("i9_", j)]]
i9[i9 < 0] <- NA
el[[paste0("amt_alamana_", j)]] <- ifelse(i3 == 3, i9, 0)
el[[paste0("amt_oformal_", j)]] <- ifelse(i3 %in% c(1,2,15), i9, 0)
el[[paste0("amt_informal_", j)]] <- ifelse(i3 %in% 7:14, i9, 0)
el[[paste0("amt_oamc_", j)]] <- ifelse(i3 %in% c(4,5,6), i9, 0)
}
el <- el %>% mutate(
loansamt_alamana = rowSums(select(., starts_with("amt_alamana_")), na.rm=TRUE),
loansamt_oformal = rowSums(select(., starts_with("amt_oformal_")), na.rm=TRUE),
loansamt_informal = rowSums(select(., starts_with("amt_informal_")), na.rm=TRUE),
loansamt_oamc = rowSums(select(., starts_with("amt_oamc_")), na.rm=TRUE),
loansamt_total = loansamt_alamana + loansamt_oformal +
loansamt_informal + loansamt_oamc
)
cat("Taux d'emprunt Al Amana (traités):",
round(mean(el$borrowed_alamana[el$treated==1], na.rm=TRUE), 3), "\n")
## Taux d'emprunt Al Amana (traités): 0.105
cat("Taux d'emprunt Al Amana (contrôles):",
round(mean(el$borrowed_alamana[el$treated==0], na.rm=TRUE), 3), "\n")
## Taux d'emprunt Al Amana (contrôles): 0.02
Activités
d’auto-emploi
# Types d'activité : Section D (d2_j = secteur : 1=agri, 2=élevage, 3-6=commerce/services)
el <- el %>% mutate(
act_agri = as.integer(rowSums(sapply(1:6, function(j) el[[paste0("d2_",j)]] == 1), na.rm=TRUE) > 0),
act_livestock = as.integer(rowSums(sapply(1:6, function(j) el[[paste0("d2_",j)]] == 2), na.rm=TRUE) > 0),
act_business = as.integer(rowSums(sapply(1:6, function(j) el[[paste0("d2_",j)]] %in% 3:6), na.rm=TRUE) > 0),
self_empl = as.integer(act_agri == 1 | act_livestock == 1 | act_business == 1)
)
# Nombre d'activités
el$act_number <- rowSums(sapply(1:6, function(j) {
nom <- el[[paste0("d1_nom", j)]]
sec <- el[[paste0("d2_", j)]]
(!is.na(nom) & nom != "" & nom != "." & !sec %in% c(7,8, NA))
}), na.rm = TRUE)
Profits et
revenus
# --- AGRICULTURE : production = ventes + autoconsommation + stock ---
# Le do-file agrège sur ~20 cultures (e20, e21, e22, e23, e19)
# Simplification : on prend les variables d'output agricole disponibles
# Ventes agricoles : e20_j * e21_j (quantité * prix vente)
sale_cereal <- rep(0, nrow(el))
for (j in 1:20) {
q_sold <- el[[paste0("e20_", j)]]; q_sold[is.na(q_sold) | q_sold < 0] <- 0
p_sold <- el[[paste0("e21_", j)]]; p_sold[is.na(p_sold) | p_sold <= 0] <- NA
q_auto <- el[[paste0("e22_", j)]]; q_auto[is.na(q_auto) | q_auto < 0] <- 0
p_auto <- el[[paste0("e23_", j)]]; p_auto[is.na(p_auto) | p_auto <= 0] <- NA
# Utiliser prix médian de l'échantillon si prix manquant
med_p <- median(ifelse(el[[paste0("e21_",j)]] > 0, el[[paste0("e21_",j)]], NA), na.rm=TRUE)
p_sold[is.na(p_sold)] <- med_p
p_auto[is.na(p_auto)] <- med_p
sale_cereal <- sale_cereal + q_sold * replace_na(p_sold, 0) +
q_auto * replace_na(p_auto, 0)
}
el$sale_agri <- sale_cereal
# Dépenses agricoles : e44 (main d'oeuvre), inputs
# Simplifié : on prend les dépenses de travail salarié agricole
el$expense_agri <- 0
for (j in 1:11) {
v <- el[[paste0("e42_", j)]]; v[is.na(v) | v < 0] <- 0
el$expense_agri <- el$expense_agri + v
}
# Stock agricole (savings_agri) : simplifié via e19 (stock restant)
el$savings_agri <- 0
for (j in 1:20) {
v <- el[[paste0("e19_", j)]]; v[is.na(v) | v < 0] <- 0
el$savings_agri <- el$savings_agri + v
}
el$prod_agri <- el$sale_agri + el$savings_agri
# --- ÉLEVAGE ---
# Ventes élevage — à adapter selon les vrais noms
el$sale_livestock <- 0
el$expense_livestock <- 0
# Dépenses élevage : f26_j (aliments) + f27_j (vétérinaire)
el$expense_livestock <- 0
for (j in 1:8) {
for (vn in c("f26_", "f27_")) {
v <- el[[paste0(vn, j)]]
if (!is.null(v)) { v[is.na(v) | v < 0] <- 0; el$expense_livestock <- el$expense_livestock + v }
}
}
# Actifs élevage — simplifié (f14/f15 non disponibles)
el$asset_livestock <- 0 # sera affiné quand les vrais noms sont connus
# --- COMMERCE/SERVICES ---
el$sale_business <- 0
for (j in 1:6) {
v <- el[[paste0("g11_", j)]]; if (!is.null(v)) { v[is.na(v)|v<0] <- 0; el$sale_business <- el$sale_business + v }
}
el$expense_business <- 0
for (j in 1:6) {
v <- el[[paste0("g22_", j)]]; if (!is.null(v)) { v[is.na(v)|v<0] <- 0; el$expense_business <- el$expense_business + v }
}
el$asset_business <- 0
for (j in 1:9) {
v <- el[[paste0("g7_", j)]]; if (!is.null(v)) { v[is.na(v)|v<0] <- 0; el$asset_business <- el$asset_business + v }
}
# --- AGRÉGATS ---
el <- el %>% mutate(
output_total = prod_agri + sale_livestock + sale_business,
assets_total = asset_livestock + asset_business,
expense_total = expense_agri + expense_livestock + expense_business,
profit_agri = prod_agri - expense_agri,
profit_livestock = sale_livestock - expense_livestock,
profit_business = sale_business - expense_business,
profit_total = profit_agri + profit_livestock + profit_business
)
cat("Profit moyen (contrôle, MAD):", round(mean(el$profit_total[el$treated==0], na.rm=TRUE), 0), "\n")
## Profit moyen (contrôle, MAD): 3144
Revenus
salariaux
# Revenus du travail salarié : Section H2, variables h11_j et h11c_j
el$income_dep <- 0
for (j in 1:22) {
v1 <- el[[paste0("h11_", j)]]; if (!is.null(v1)) { v1[is.na(v1)|v1<0] <- 0; el$income_dep <- el$income_dep + v1 }
v2 <- el[[paste0("h11c", j)]]; if (!is.null(v2)) { v2[is.na(v2)|v2<0] <- 0; el$income_dep <- el$income_dep + v2 }
}
el <- el %>% mutate(
income_work = profit_total + income_dep,
income_total = income_work
)
cat("Revenu salarial moyen (contrôle):", round(mean(el$income_dep[el$treated==0], na.rm=TRUE), 0), "\n")
## Revenu salarial moyen (contrôle): 15618
Consommation
# Consommation alimentaire : h1_j et h2_j (par semaine * 4.345)
el$cons_food <- 0
for (j in c(1:14, 17, 18)) {
v1 <- el[[paste0("h1_", j)]]; if (!is.null(v1)) { v1[is.na(v1)|v1<0] <- 0; el$cons_food <- el$cons_food + v1 }
v2 <- el[[paste0("h2_", j)]]; if (!is.null(v2)) { v2[is.na(v2)|v2<0] <- 0; el$cons_food <- el$cons_food + v2 }
}
el$cons_food <- el$cons_food * 4.345
# Durables : Section C (c4_j / 12)
el$cons_durables <- 0
for (j in 1:31) {
v <- el[[paste0("c4_", j)]]; if (!is.null(v)) { v[is.na(v)|v<0] <- 0; el$cons_durables <- el$cons_durables + v }
}
el$cons_durables <- el$cons_durables / 12
# Consommation de fêtes : h4_5 (ramadan) + h4_6 (Aïd) + h3_11 / 12
el$cons_ramadan <- ifelse(!is.na(el$h4_5) & el$h4_5 >= 0, el$h4_5 / 12, 0)
el$cons_aid <- ifelse(!is.na(el$h4_6) & el$h4_6 >= 0, el$h4_6 / 12, 0)
el$cons_social <- el$cons_ramadan + el$cons_aid
if ("h3_11" %in% names(el)) el$cons_social <- el$cons_social + ifelse(!is.na(el$h3_11) & el$h3_11 >= 0, el$h3_11, 0)
# Santé : h3_8
el$cons_health <- ifelse(!is.na(el$h3_8) & el$h3_8 >= 0, el$h3_8, 0)
# Éducation : h4_1 / 12
el$cons_schooling <- ifelse(!is.na(el$h4_1) & el$h4_1 >= 0, el$h4_1 / 12, 0)
# Non-durables (inclut food + autres postes hebdo + mensuels)
el$cons_nondurables <- el$cons_food
for (j in c(15, 16, 19, 20, 21)) {
v1 <- el[[paste0("h1_", j)]]; if (!is.null(v1)) { v1[is.na(v1)|v1<0] <- 0; el$cons_nondurables <- el$cons_nondurables + v1 * 4.345 }
v2 <- el[[paste0("h2_", j)]]; if (!is.null(v2)) { v2[is.na(v2)|v2<0] <- 0; el$cons_nondurables <- el$cons_nondurables + v2 * 4.345 }
}
for (j in c(1:9, 11:17)) {
v <- el[[paste0("h3_", j)]]; if (!is.null(v)) { v[is.na(v)|v<0] <- 0; el$cons_nondurables <- el$cons_nondurables + v }
}
for (j in 1:10) {
v <- el[[paste0("h4_", j)]]; if (!is.null(v)) { v[is.na(v)|v<0] <- 0; el$cons_nondurables <- el$cons_nondurables + v / 12 }
}
# Consommation totale
el$consumption <- el$cons_durables + el$cons_nondurables
el$consumption[el$cons_food == 0 | is.na(el$cons_food)] <- NA
cat("Consommation moyenne (contrôle):", round(mean(el$consumption[el$treated==0], na.rm=TRUE), 0), "MAD/mois\n")
## Consommation moyenne (contrôle): 3004 MAD/mois
Heures travaillées (7
derniers jours)
# Waves 2, 3, 4 : h05_ai_j (heures/jour) * h04_ai_j (jours) par activité i, membre j
# h03_ri_j = type de travail (5 ou 7 = auto-emploi, 1-4,6,8-9 = extérieur)
# Initialisation
el$hours_self_age6_65 <- 0
el$hours_outside_age6_65 <- 0
el$hours_chores_age6_65 <- 0
for (j in 1:24) {
age_var <- el[[paste0("a7_", j)]]
in_group <- !is.na(age_var) & age_var >= 6 & age_var <= 65
for (i in 1:3) {
type_var <- el[[paste0("h03_r", i, "_", j)]]
hrs_var <- el[[paste0("h05_a", i, "_", j)]]
days_var <- el[[paste0("h04_a", i, "_", j)]]
if (!is.null(type_var) & !is.null(hrs_var) & !is.null(days_var)) {
hrs_var[is.na(hrs_var) | hrs_var < 0 | hrs_var >= 18] <- 0
days_var[is.na(days_var) | days_var < 0 | days_var > 7] <- 0
contrib <- hrs_var * days_var
is_self <- !is.na(type_var) & type_var %in% c(5, 7)
is_outside <- !is.na(type_var) & (type_var %in% c(1:4, 6, 8, 9))
el$hours_self_age6_65 <- el$hours_self_age6_65 + ifelse(in_group & is_self, contrib, 0)
el$hours_outside_age6_65 <- el$hours_outside_age6_65 + ifelse(in_group & is_outside, contrib, 0)
}
}
# Tâches domestiques : h07_j * h08_j
chores_days <- el[[paste0("h07_", j)]]; chores_hrs <- el[[paste0("h08_", j)]]
if (!is.null(chores_days) & !is.null(chores_hrs)) {
chores_days[is.na(chores_days)|chores_days<0|chores_days>7] <- 0
chores_hrs[is.na(chores_hrs)|chores_hrs<0|chores_hrs>=18] <- 0
el$hours_chores_age6_65 <- el$hours_chores_age6_65 + ifelse(in_group, chores_days * chores_hrs, 0)
}
}
# Pour wave 1 : a12_pj (auto-emploi) + a14_j (extérieur) + a13_j (tâches dom.)
for (j in 1:11) {
age_var <- el[[paste0("a7_", j)]]
in_group <- !is.na(age_var) & age_var >= 6 & age_var <= 65 & el$wave == 1
sp <- el[[paste0("a12_p", j)]]; sp[is.na(sp)|sp<0|sp>=126] <- 0
sm <- el[[paste0("a12_m", j)]]; sm[is.na(sm)|sm<0|sm>=126] <- 0
ou <- el[[paste0("a14_", j)]]; ou[is.na(ou)|ou<0|ou>=126] <- 0
ch <- el[[paste0("a13_", j)]]; ch[is.na(ch)|ch<0|ch>=126] <- 0
el$hours_self_age6_65 <- el$hours_self_age6_65 + ifelse(in_group, sp + sm, 0)
el$hours_outside_age6_65 <- el$hours_outside_age6_65 + ifelse(in_group, ou, 0)
el$hours_chores_age6_65 <- el$hours_chores_age6_65 + ifelse(in_group, ch, 0)
}
el$hours_total_age6_65 <- el$hours_self_age6_65 + el$hours_outside_age6_65 + el$hours_chores_age6_65
# Mettre à NA si aucune donnée sur les heures (nonmiss = 0)
el$hours_total_age6_65[el$hours_total_age6_65 == 0 & el$wave != 1] <- NA
cat("Heures totales moy. (contrôle):", round(mean(el$hours_total_age6_65[el$treated==0], na.rm=TRUE), 1), "\n")
## Heures totales moy. (contrôle): 146.4
cat("Dont extérieur (contrôle):", round(mean(el$hours_outside_age6_65[el$treated==0], na.rm=TRUE), 1), "\n")
## Dont extérieur (contrôle): 33.4
Scolarisation et
autonomie des femmes
# Scolarisation 6-15 : a15_j == 1 (en école) si a7_j in [6,15]
el$kids_inschool <- 0
el$kids_6_15 <- 0
for (j in 1:24) {
age <- el[[paste0("a7_", j)]]; sch <- el[[paste0("a15_", j)]]
if (!is.null(age) & !is.null(sch)) {
in_range <- !is.na(age) & age >= 6 & age < 16
el$kids_inschool <- el$kids_inschool + ifelse(in_range & !is.na(sch) & sch == 1, 1, 0)
el$kids_6_15 <- el$kids_6_15 + ifelse(in_range, 1, 0)
}
}
el$sharekids_inschool <- ifelse(el$kids_6_15 > 0, el$kids_inschool / el$kids_6_15, 0)
# Adolescents 16-20
el$teens_inschool <- 0
el$members_16_20 <- 0
for (j in 1:24) {
age <- el[[paste0("a7_", j)]]; sch <- el[[paste0("a11_", j)]]; act <- el[[paste0("a10_", j)]]
if (!is.null(age)) {
in_range <- !is.na(age) & age >= 16 & age <= 20
el$members_16_20 <- el$members_16_20 + ifelse(in_range, 1, 0)
if (!is.null(sch) & !is.null(act))
el$teens_inschool <- el$teens_inschool + ifelse(in_range & !is.na(sch) & sch == 13 & !is.na(act) & act != 2, 1, 0)
}
}
el$shareteens_inschool <- ifelse(el$members_16_20 > 0, el$teens_inschool / el$members_16_20, 0)
# Index autonomie femmes : j1 à j9, j9b, j10 à j13
women_vars <- list()
for (i in 1:9) {
w <- rep(0, nrow(el))
for (j in 1:11) { v <- el[[paste0("j", i, "_", j)]]; if(!is.null(v)) w <- w + ifelse(!is.na(v) & v == 1, 1, 0) }
women_vars[[i]] <- as.integer(w > 0)
}
if ("j9b1" %in% names(el)) {
w10 <- rep(0, nrow(el))
for (j in 1:11) { v <- el[[paste0("j9b", j)]]; if(!is.null(v)) w10 <- w10 + ifelse(!is.na(v) & v == 1, 1, 0) }
women_vars[[10]] <- as.integer(w10 > 0)
} else { women_vars[[10]] <- rep(0, nrow(el)) }
for (k in 10:13) {
v <- el[[paste0("j", k)]]
women_vars[[k]] <- if (!is.null(v)) as.integer(!is.na(v) & v %in% c(3, 4)) else rep(0, nrow(el))
}
# Standardiser et sommer pour l'index
women_mat <- do.call(cbind, women_vars)
women_std <- scale(women_mat)
el$women_index <- rowSums(women_std, na.rm=TRUE)
# Activités gérées par des femmes
el$women_act_number <- 0
for (j in 1:6) {
resp <- el[[paste0("d3_", j)]]
if (!is.null(resp)) {
for (i in 1:24) {
sex <- el[[paste0("a4_", i)]]
if (!is.null(sex)) {
el$women_act_number <- el$women_act_number +
ifelse(!is.na(resp) & resp == i & !is.na(sex) & sex == 2, 1, 0)
}
}
}
}
el$women_act <- as.integer(el$women_act_number > 0)
Variables de contrôle
et taille du ménage
# Chef de ménage : a3_j == 1 (chef), a4_j (sexe : 1=homme), a7_j (âge)
el$head_male <- 0
el$head_age <- NA
for (j in 1:16) {
is_head <- !is.na(el[[paste0("a3_", j)]]) & el[[paste0("a3_", j)]] == 1
sex_var <- el[[paste0("a4_", j)]]
age_var <- el[[paste0("a7_", j)]]
if (!is.null(sex_var)) el$head_male <- ifelse(is_head & !is.na(sex_var) & sex_var == 1, 1, el$head_male)
if (!is.null(age_var)) el$head_age <- ifelse(is_head & !is.na(age_var) & age_var >= 0, age_var, el$head_age)
}
# Membres résidents
el$members_resid <- 0
el$nadults_resid <- 0
el$nchildren_resid <- 0
for (j in 1:17) {
a3 <- el[[paste0("a3_", j)]]; a5 <- el[[paste0("a5_", j)]]; a7 <- el[[paste0("a7_", j)]]
if (!is.null(a3) & !is.null(a5) & !is.null(a7)) {
is_resid <- (!is.na(a5) & a5 %in% c(3,5)) | (!is.na(a5) & !a5 %in% c(3,5) & !is.na(a3) & a3 == 1)
el$members_resid <- el$members_resid + ifelse(is_resid, 1, 0)
el$nadults_resid <- el$nadults_resid + ifelse(is_resid & !is.na(a7) & a7 >= 16, 1, 0)
el$nchildren_resid <- el$nchildren_resid + ifelse(is_resid & !is.na(a7) & a7 >= 0 & a7 < 16, 1, 0)
}
}
cat("Membres résidents moy.:", round(mean(el$members_resid, na.rm=TRUE), 2), "\n")
## Membres résidents moy.: 6.11
Échantillon haute
probabilité et score de propension
# Le fichier contient score1, score2, score3 et client_admin
# On utilise score3 (le plus récent) ou client comme proxy
# Définir l'échantillon "haute probabilité" :
# Dans le papier, c'est le top quartile du score de propension
# On utilise score3 si disponible, sinon score1
score_var <- if ("score3" %in% names(el)) "score3" else if ("score1" %in% names(el)) "score1" else NULL
if (!is.null(score_var)) {
q75 <- quantile(el[[score_var]], 0.75, na.rm=TRUE)
el$highp <- as.integer(el[[score_var]] >= q75)
cat("N haute probabilité:", sum(el$highp, na.rm=TRUE), "\n")
} else {
# Fallback : tous les ménages
el$highp <- 1
cat("Score non trouvé, on utilise tous les ménages\n")
}
## N haute probabilité: 1388
# Covariables de contrôle disponibles (traduit du papier)
controls_ok <- intersect(
c("members_resid", "nadults_resid", "head_age",
"act_livestock", "act_business",
"cm_resp_activ", "ccm_resp_activ"),
names(el)
)
cat("Covariables disponibles:", paste(controls_ok, collapse=", "), "\n")
## Covariables disponibles: members_resid, nadults_resid, head_age, act_livestock, act_business
Fonctions d’inférence
par randomisation
# Fonction principale : calcule coef OLS + p-value RI + IC 95%
ri_test <- function(formula, data, treat_var="treated",
pair_var="pair_id", n_perm=999) {
vars_needed <- all.vars(formula)
data_clean <- data %>% select(all_of(c(vars_needed, pair_var))) %>% drop_na()
if (nrow(data_clean) < 10) return(list(beta=NA, p_ri=NA, ci_lo=NA, ci_hi=NA))
fit_obs <- lm(formula, data=data_clean)
beta_obs <- coef(fit_obs)[treat_var]
if (is.na(beta_obs)) return(list(beta=NA, p_ri=NA, ci_lo=NA, ci_hi=NA))
pairs <- unique(data_clean[[pair_var]])
beta_perm <- numeric(n_perm)
for (i in seq_len(n_perm)) {
dp <- data_clean
for (p in pairs) {
idx <- which(dp[[pair_var]] == p)
if (length(idx) >= 2) dp[[treat_var]][idx] <- sample(dp[[treat_var]][idx])
}
beta_perm[i] <- coef(lm(formula, data=dp))[treat_var]
}
p_ri <- mean(abs(beta_perm) >= abs(beta_obs))
ci_lo <- beta_obs - quantile(abs(beta_perm), 0.975)
ci_hi <- beta_obs + quantile(abs(beta_perm), 0.975)
list(beta=beta_obs, p_ri=p_ri, ci_lo=ci_lo, ci_hi=ci_hi)
}
build_formula <- function(outcome, treat_var="treated",
pair_var="pair_id", covariates=NULL) {
rhs <- treat_var
if (!is.null(covariates) && length(covariates) > 0)
rhs <- paste(c(rhs, covariates), collapse=" + ")
as.formula(paste(outcome, "~", rhs, "+", pair_var))
}
make_row <- function(ri, label, ctrl_mean, n_obs, digs=0) {
if (is.na(ri$beta)) return(tibble(Variable=label, Coefficient=NA, ` `="", `p-val RI`=NA,
`IC 95% bas`=NA, `IC 95% haut`=NA, `Moy. contrôle`=round(ctrl_mean,digs), N=n_obs))
stars <- ifelse(ri$p_ri<0.01,"***", ifelse(ri$p_ri<0.05,"**", ifelse(ri$p_ri<0.10,"*","")))
tibble(Variable=label, Coefficient=round(ri$beta,digs), ` `=stars,
`p-val RI`=round(ri$p_ri,3), `IC 95% bas`=round(ri$ci_lo,digs),
`IC 95% haut`=round(ri$ci_hi,digs), `Moy. contrôle`=round(ctrl_mean,digs), N=n_obs)
}
run_table <- function(var_list, sample_data, cov, n_perm=999, digs=0) {
var_ok <- var_list[unlist(var_list) %in% names(sample_data)]
if (length(var_ok)==0) { message("Aucune variable trouvée."); return(tibble()) }
map_dfr(names(var_ok), function(label) {
var <- var_ok[[label]]
dat <- sample_data %>% select(all_of(c(var,"treated","pair_id",cov))) %>% drop_na()
ctrl_mean <- mean(dat[[var]][dat$treated==0], na.rm=TRUE)
ri <- ri_test(build_formula(var, covariates=cov), dat, n_perm=n_perm)
make_row(ri, label, ctrl_mean, nrow(dat), digs)
})
}
cat("Fonctions RI prêtes.\n")
## Fonctions RI prêtes.
Table 9 — Impact sur
les emprunteurs (IV/LATE)
# Instrument : village traité → emprunt Al Amana
# LATE = ITT / first stage ≈ ITT / 0.17
key_outcomes <- list(
"Actifs totaux (MAD)" = "assets_total",
"Production totale (MAD)" = "output_total",
"Dépenses totales (MAD)" = "expense_total",
"Profit total (MAD)" = "profit_total",
"Revenu salarial (MAD)" = "income_dep",
"Heures extérieur" = "hours_outside_age6_65",
"Consommation (MAD/mois)" = "consumption"
)
key_ok <- key_outcomes[unlist(key_outcomes) %in% names(el_hp)]
if ("client_admin" %in% names(el_hp) && length(key_ok) > 0) {
cov_hp <- cov_ok[cov_ok %in% names(el_hp)]
panelB_iv <- map_dfr(names(key_ok), function(label) {
var <- key_ok[[label]]
dat <- el_hp %>% select(all_of(c(var,"client_admin","treated","pair_id",cov_hp))) %>% drop_na()
ctrl_mean <- mean(dat[[var]][dat$treated==0], na.rm=TRUE)
cov_str <- if (length(cov_hp)>0) paste(cov_hp, collapse=" + ") else "1"
form_iv <- as.formula(paste(var, "~ client_admin +", cov_str, "+ pair_id |",
"treated +", cov_str, "+ pair_id"))
fit_iv <- tryCatch(ivreg(form_iv, data=dat), error=function(e) NULL)
if (is.null(fit_iv)) return(make_row(list(beta=NA,p_ri=NA,ci_lo=NA,ci_hi=NA), label, ctrl_mean, nrow(dat)))
beta_obs <- coef(fit_iv)["client_admin"]
pairs <- unique(dat$pair_id)
bp <- numeric(999)
for (i in 1:999) {
dp <- dat
for (p in pairs) { idx <- which(dp$pair_id==p); if(length(idx)>=2) dp$treated[idx] <- sample(dp$treated[idx]) }
fp <- tryCatch(ivreg(form_iv, data=dp), error=function(e) NULL)
bp[i] <- if (!is.null(fp)) coef(fp)["client_admin"] else NA_real_
}
bp <- bp[!is.na(bp)]
p_ri <- mean(abs(bp) >= abs(beta_obs))
ci_lo <- beta_obs - quantile(abs(bp), 0.975)
ci_hi <- beta_obs + quantile(abs(bp), 0.975)
make_row(list(beta=beta_obs, p_ri=p_ri, ci_lo=ci_lo, ci_hi=ci_hi), label, ctrl_mean, nrow(dat))
})
kable(panelB_iv, caption="Table 9 — IV/LATE — p-values par RI") %>%
kable_styling(bootstrap_options=c("striped","hover","condensed"), full_width=FALSE) %>%
footnote(general=paste("* p<0.10, ** p<0.05, *** p<0.01 (RI, 999 permutations).",
"Instrument : village traité. LATE = effet causal sur les compliers.", sep="\n"))
} else {
cat("client_admin ou variables d'outcome non disponibles.\n")
}
Table 9 — IV/LATE — p-values par RI
|
Variable
|
Coefficient
|
|
p-val RI
|
IC 95% bas
|
IC 95% haut
|
Moy. contrôle
|
N
|
|
Actifs totaux (MAD)
|
-467
|
|
1.000
|
-145079
|
144145
|
73
|
1365
|
|
Production totale (MAD)
|
18204
|
|
0.874
|
-1735862
|
1772270
|
5195
|
1365
|
|
Dépenses totales (MAD)
|
1474
|
|
0.901
|
-152543
|
155490
|
550
|
1365
|
|
Profit total (MAD)
|
16730
|
|
0.874
|
-2016641
|
2050102
|
4644
|
1365
|
|
Revenu salarial (MAD)
|
-10281
|
|
0.886
|
-1618103
|
1597541
|
18972
|
1365
|
|
Heures extérieur
|
-17
|
|
0.935
|
-2755
|
2722
|
37
|
1365
|
|
Consommation (MAD/mois)
|
-2380
|
|
0.932
|
-455823
|
451062
|
3257
|
1364
|
|
Note:
|
* p<0.10, ** p<0.05, *** p<0.01 (RI, 999
permutations).
Instrument : village traité. LATE = effet causal sur
les compliers.
|