H₅: Organizational practices

Code

library(tidyverse)
library(targets)
library(tidybayes)
library(patchwork)
library(scales)
library(ggh4x)
library(glue)
library(gt)
library(gtExtras)

tar_config_set(
  store = here::here("_targets"),
  script = here::here("_targets.R")
)

preds_all <- tar_read(preds_conditional_treatment_only)

invisible(list2env(tar_read(graphic_functions), .GlobalEnv))
invisible(list2env(tar_read(table_functions), .GlobalEnv))

theme_set(theme_ngo())

H_5a: Financial transparency

Hypothesis 5_a: Financial transparency

Donors will show increased willingness to donate to NGOs that are financially transparent

Mechanism: Perception of efficacy

Estimand

\[ \begin{aligned} \theta =&\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Transparency} = \text{Yes} \right) \bigr] \end{aligned} \]

Code

h5a_mms <- preds_all %>% 
  group_by(feat_transp_short, .draw) %>% 
  summarize(avg = mean(.epred))

h5a_amces <- h5a_mms %>% 
  group_by(feat_transp_short) %>% 
  compare_levels(variable = avg, by = feat_transp_short, comparison = "control")

Code

p_h5a_mms <- h5a_mms %>% 
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_percent()) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  labs(
    x = "Overall average predicted probability",
    y = NULL,
    fill = NULL,
    title = "Estimated marginal means"
  )

p_h5a_amces <- h5a_amces %>% 
  ungroup() %>% 
  separate_wider_delim(
    feat_transp_short,
    delim = " - ", 
    names = c("feat_transp_short", "reference_level")
  ) %>% 
  add_row(avg = 0, feat_transp_short = unique(.$reference_level)) %>%
  mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5a_mms$feat_transp_short))) %>% 
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_pp) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  labs(
    x = "Percentage point change in probability of choice selection",
    y = NULL,
    fill = NULL,
    title = "Average marginal component effects (AMCEs)"
  )

p_h5a_mms | p_h5a_amces

Code

h5a_tbl_mm <- h5a_mms %>% 
  group_by(feat_transp_short) %>% 
  median_qi() %>% 
  mutate(nice = glue("{fmt_decimal(avg)}<br>{build_ci(.lower, .upper)}")) %>% 
  arrange(desc(feat_transp_short)) %>% 
  select(feat_transp_short, nice)

h5a_tbl_amces <- h5a_amces %>% 
  group_by(feat_transp_short) %>% 
  summarize(
    median_qi(avg),
    p_gt_0 = sum(avg > 0) / n()
  ) %>%
  mutate(p_neq_0 = ifelse(y >= 0, p_gt_0, 1 - p_gt_0)) %>% 
  mutate(nice = glue("{fmt_decimal(y)}<br>{build_ci(ymin, ymax)}")) %>% 
  mutate(feat_transp_short = str_replace(feat_transp_short, " - ", "−<br>")) %>% 
  mutate(across(starts_with("p_"), ~fmt_decimal(.))) %>% 
  arrange(desc(feat_transp_short)) %>% 
  select(contrast = feat_transp_short, amce_nice = nice, p_neq_0)

bind_cols(
  h5a_tbl_mm, 
  add_row(h5a_tbl_amces, contrast = "*(Reference)*")
) %>% 
  gt() %>% 
  sub_missing(columns = everything(), missing_text = "—") %>% 
  fmt_markdown(columns = c(nice, amce_nice, contrast)) %>%
  cols_align(align = "center", columns = everything()) %>% 
  cols_align(align = "left", columns = c(feat_transp_short, contrast)) %>% 
  cols_label(
    feat_transp_short = "Transparency", 
    nice = "Posterior EMM",
    contrast = "Contrast",
    amce_nice = "Posterior AMCE",
    p_neq_0 = "*p*~direction~",
    .fn = md
  ) %>% 
  tab_style(
    style = cell_text(v_align = "top"),
    locations = cells_body()
  ) %>% 
  tab_footnote(
    footnote = "Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.",
    locations = cells_column_labels(columns = c(nice, amce_nice))
  ) %>% 
  tab_footnote(
    footnote = md("The probability of direction (*p*~direction~) is the probability that the posterior AMCE is strictly positive or negative—it is the proportion of the posterior AMCE that is the sign of the median."),
    locations = cells_column_labels(columns = p_neq_0)
  ) %>% 
  gt_add_divider(columns = nice, style = "dashed", weight = px(1)) %>% 
  opt_footnote_marks(marks = "standard") %>% 
  opt_horizontal_padding(3) %>% 
  opts_theme()

Transparency	Posterior EMM^*	Contrast	Posterior AMCE^*	p_direction^†
Yes	0.307 [0.299, 0.314]	Yes− No	0.103 [0.094, 0.112]	1.000
No	0.204 [0.198, 0.210]	(Reference)	—	—
^* Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.
^† The probability of direction (p_direction) is the probability that the posterior AMCE is strictly positive or negative—it is the proportion of the posterior AMCE that is the sign of the median.

H_5b: Relationship with host government and organizational practices

Transparency

Hypothesis 5_b: Relationship with host government, organizational practices

Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are also financially transparent

\[ \begin{aligned} \theta =&\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Transparency} = \text{Yes} \mid \text{Relationship = Under crackdown or Criticized} \right) \bigr] - \\ &\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Transparency} = \text{Yes} \mid \text{Relationship = Friendly} \right) \bigr] \end{aligned} \]

Code

h5b_mms <- preds_all %>% 
  group_by(feat_transp_short, feat_govt, .draw) %>% 
  summarize(avg = mean(.epred))

h5b_amces <- h5b_mms %>% 
  group_by(feat_govt) %>% 
  compare_levels(variable = avg, by = feat_transp_short, comparison = "control")

Code

p_h5b_mms <- h5b_mms %>% 
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_percent()) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  facet_wrap(vars(feat_govt), ncol = 1) +
  labs(
    x = "Overall average predicted probability",
    y = NULL,
    fill = NULL,
    title = "Estimated marginal means"
  ) +
  theme(
    legend.position = "top",
    legend.justification = "left",
    legend.margin = margin(l = -7, t = 0)
  )

p_h5b_amces <- h5b_amces %>% 
  ungroup() %>% 
  separate_wider_delim(
    feat_transp_short,
    delim = " - ", 
    names = c("feat_transp_short", "reference_level")
  ) %>% 
  add_row(avg = 0, feat_transp_short = "No", feat_govt = "Friendly relationship with government") %>%
  add_row(avg = 0, feat_transp_short = "No", feat_govt = "Criticized by government") %>%
  add_row(avg = 0, feat_transp_short = "No", feat_govt = "Under government crackdown") %>%
  mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5b_mms$feat_transp_short))) %>% 
  mutate(feat_govt = factor(feat_govt, levels = levels(h5b_mms$feat_govt))) %>% 
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_pp) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  facet_wrap(vars(feat_govt), ncol = 1) +
  labs(
    x = "Percentage point change in probability of choice selection",
    y = NULL,
    fill = NULL,
    title = "Difference in estimated marginal means"
  )

p_h5b_mms | p_h5b_amces

Code

h5b_tbl_mm <- h5b_mms %>% 
  group_by(feat_govt, feat_transp_short) %>% 
  median_qi() %>% 
  mutate(nice = glue("{fmt_decimal(avg)}<br>{build_ci(.lower, .upper)}")) %>% 
  arrange(desc(feat_govt), desc(feat_transp_short)) %>% 
  select(feat_transp_short, feat_govt, nice)

h5b_tbl_amces <- h5b_amces %>% 
  group_by(feat_govt, feat_transp_short) %>% 
  summarize(
    median_qi(avg),
    p_gt_0 = sum(avg > 0) / n()
  ) %>%
  ungroup() %>% 
  mutate(p_neq_0 = ifelse(y >= 0, p_gt_0, 1 - p_gt_0)) %>% 
  mutate(nice = glue("{fmt_decimal(y)}<br>{build_ci(ymin, ymax)}")) %>% 
  mutate(feat_transp_short = str_replace(feat_transp_short, " - ", "−<br>")) %>% 
  mutate(across(starts_with("p_"), ~fmt_decimal(.))) %>% 
  arrange(desc(feat_govt)) %>% 
  select(contrast = feat_transp_short, diff_nice = nice, p_neq_0)

bind_cols(
  h5b_tbl_mm, 
  h5b_tbl_amces %>% 
    add_row(contrast = NA, .after = 1) %>% 
    add_row(contrast = NA, .after = 3) %>% 
    add_row(contrast = NA, .after = 5)
) %>% 
  mutate(feat_govt = fct_relabel(feat_govt, ~paste("Relationship with government:", .x))) %>% 
  group_by(feat_govt) %>% 
  gt() %>% 
  sub_missing(columns = everything(), missing_text = "—") %>% 
  fmt_markdown(columns = c(nice, diff_nice, contrast)) %>%
  cols_align(align = "center", columns = everything()) %>% 
  cols_align(align = "left", columns = c(feat_transp_short, contrast)) %>% 
  cols_label(
    feat_transp_short = "Transparency", 
    nice = "Posterior EMM",
    contrast = "Contrast",
    diff_nice = "Posterior ∆",
    p_neq_0 = "*p*~direction~",
    .fn = md
  ) %>% 
  tab_style(
    style = cell_text(v_align = "top"),
    locations = cells_body()
  ) %>% 
  tab_style(
    style = cell_fill(color = "grey90"),
    locations = cells_row_groups()
  ) %>% 
  tab_footnote(
    footnote = "Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.",
    locations = cells_column_labels(columns = c(nice, diff_nice))
  ) %>% 
  tab_footnote(
    footnote = md("The probability of direction (*p*~direction~) is the probability that the posterior difference in marginal means is strictly positive or negative—it is the proportion of the posterior difference in marginal means that is the sign of the median."),
    locations = cells_column_labels(columns = p_neq_0)
  ) %>% 
  gt_add_divider(columns = nice, style = "dashed", weight = px(1)) %>% 
  opt_footnote_marks(marks = "standard") %>% 
  opt_horizontal_padding(3) %>% 
  opts_theme()

Transparency	Posterior EMM^*	Contrast	Posterior ∆^*	p_direction^†
Relationship with government: Under government crackdown
Yes	0.254 [0.245, 0.264]	Yes− No	0.091 [0.083, 0.098]	1.000
No	0.163 [0.156, 0.171]	—	—	—
Relationship with government: Criticized by government
Yes	0.294 [0.284, 0.304]	Yes− No	0.101 [0.092, 0.110]	1.000
No	0.193 [0.185, 0.201]	—	—	—
Relationship with government: Friendly relationship with government
Yes	0.372 [0.361, 0.383]	Yes− No	0.118 [0.108, 0.128]	1.000
No	0.254 [0.245, 0.264]	—	—	—
^* Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.
^† The probability of direction (p_direction) is the probability that the posterior difference in marginal means is strictly positive or negative—it is the proportion of the posterior difference in marginal means that is the sign of the median.

Accountability

Hypothesis 5_b2: Relationship with host government, organizational practices

Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are accountable and hold regular third party audits

\[ \begin{aligned} \theta =&\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Accountability} = \text{Yes} \mid \text{Relationship = Under crackdown or Criticized} \right) \bigr] - \\ &\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Accountability} = \text{Yes} \mid \text{Relationship = Friendly} \right) \bigr] \end{aligned} \]

Code

h5b2_mms <- preds_all %>% 
  group_by(feat_acc_short, feat_govt, .draw) %>% 
  summarize(avg = mean(.epred))

h5b2_amces <- h5b2_mms %>% 
  group_by(feat_govt) %>% 
  compare_levels(variable = avg, by = feat_acc_short, comparison = "control")

Code

p_h5b2_mms <- h5b2_mms %>% 
  ggplot(aes(x = avg, y = feat_acc_short, fill = feat_acc_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_percent()) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(9, 4)], guide = "none") +
  facet_wrap(vars(feat_govt), ncol = 1) +
  labs(
    x = "Overall average predicted probability",
    y = NULL,
    fill = NULL,
    title = "Estimated marginal means"
  ) +
  theme(
    legend.position = "top",
    legend.justification = "left",
    legend.margin = margin(l = -7, t = 0)
  )

p_h5b2_amces <- h5b2_amces %>% 
  ungroup() %>% 
  separate_wider_delim(
    feat_acc_short,
    delim = " - ", 
    names = c("feat_acc_short", "reference_level")
  ) %>% 
  add_row(avg = 0, feat_acc_short = "No", feat_govt = "Friendly relationship with government") %>%
  add_row(avg = 0, feat_acc_short = "No", feat_govt = "Criticized by government") %>%
  add_row(avg = 0, feat_acc_short = "No", feat_govt = "Under government crackdown") %>%
  mutate(feat_acc_short = factor(feat_acc_short, levels = levels(h5b2_mms$feat_acc_short))) %>% 
  mutate(feat_govt = factor(feat_govt, levels = levels(h5b2_mms$feat_govt))) %>% 
  ggplot(aes(x = avg, y = feat_acc_short, fill = feat_acc_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_pp) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(9, 4)], guide = "none") +
  facet_wrap(vars(feat_govt), ncol = 1) +
  labs(
    x = "Percentage point change in probability of choice selection",
    y = NULL,
    fill = NULL,
    title = "Difference in estimated marginal means"
  )

p_h5b2_mms | p_h5b2_amces

Code

h5b2_tbl_mm <- h5b2_mms %>% 
  group_by(feat_govt, feat_acc_short) %>% 
  median_qi() %>% 
  mutate(nice = glue("{fmt_decimal(avg)}<br>{build_ci(.lower, .upper)}")) %>% 
  arrange(desc(feat_govt), desc(feat_acc_short)) %>% 
  select(feat_acc_short, feat_govt, nice)

h5b2_tbl_amces <- h5b2_amces %>% 
  group_by(feat_govt, feat_acc_short) %>% 
  summarize(
    median_qi(avg),
    p_gt_0 = sum(avg > 0) / n()
  ) %>%
  ungroup() %>% 
  mutate(p_neq_0 = ifelse(y >= 0, p_gt_0, 1 - p_gt_0)) %>% 
  mutate(nice = glue("{fmt_decimal(y)}<br>{build_ci(ymin, ymax)}")) %>% 
  mutate(feat_acc_short = str_replace(feat_acc_short, " - ", "−<br>")) %>% 
  mutate(across(starts_with("p_"), ~fmt_decimal(.))) %>% 
  arrange(desc(feat_govt)) %>% 
  select(contrast = feat_acc_short, diff_nice = nice, p_neq_0)

bind_cols(
  h5b2_tbl_mm, 
  h5b2_tbl_amces %>% 
    add_row(contrast = NA, .after = 1) %>% 
    add_row(contrast = NA, .after = 3) %>% 
    add_row(contrast = NA, .after = 5)
) %>% 
  mutate(feat_govt = fct_relabel(feat_govt, ~paste("Relationship with government:", .x))) %>% 
  group_by(feat_govt) %>% 
  gt() %>% 
  sub_missing(columns = everything(), missing_text = "—") %>% 
  fmt_markdown(columns = c(nice, diff_nice, contrast)) %>%
  cols_align(align = "center", columns = everything()) %>% 
  cols_align(align = "left", columns = c(feat_acc_short, contrast)) %>% 
  cols_label(
    feat_acc_short = "Accountability", 
    nice = "Posterior EMM",
    contrast = "Contrast",
    diff_nice = "Posterior ∆",
    p_neq_0 = "*p*~direction~",
    .fn = md
  ) %>% 
  tab_style(
    style = cell_text(v_align = "top"),
    locations = cells_body()
  ) %>% 
  tab_style(
    style = cell_fill(color = "grey90"),
    locations = cells_row_groups()
  ) %>% 
  tab_footnote(
    footnote = "Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.",
    locations = cells_column_labels(columns = c(nice, diff_nice))
  ) %>% 
  tab_footnote(
    footnote = md("The probability of direction (*p*~direction~) is the probability that the posterior difference in marginal means is strictly positive or negative—it is the proportion of the posterior difference in marginal means that is the sign of the median."),
    locations = cells_column_labels(columns = p_neq_0)
  ) %>% 
  gt_add_divider(columns = nice, style = "dashed", weight = px(1)) %>% 
  opt_footnote_marks(marks = "standard") %>% 
  opt_horizontal_padding(3) %>% 
  opts_theme()

Accountability	Posterior EMM^*	Contrast	Posterior ∆^*	p_direction^†
Relationship with government: Under government crackdown
Yes	0.253 [0.244, 0.263]	Yes− No	0.089 [0.081, 0.097]	1.000
No	0.164 [0.157, 0.172]	—	—	—
Relationship with government: Criticized by government
Yes	0.293 [0.283, 0.303]	Yes− No	0.099 [0.090, 0.108]	1.000
No	0.194 [0.186, 0.202]	—	—	—
Relationship with government: Friendly relationship with government
Yes	0.371 [0.360, 0.381]	Yes− No	0.115 [0.106, 0.125]	1.000
No	0.256 [0.246, 0.265]	—	—	—
^* Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.
^† The probability of direction (p_direction) is the probability that the posterior difference in marginal means is strictly positive or negative—it is the proportion of the posterior difference in marginal means that is the sign of the median.

H_5c: Relationship with host government, organizational practices, and funding sources

Hypothesis 5_c: Relationship with host government, organizational practices, and funding sources

Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are also financially transparent and are funded primarily by numerous small private donors

Estimand

TODO

Code

h5c_mms <- preds_all %>% 
  group_by(feat_transp_short, feat_govt, feat_funding_short, .draw) %>% 
  summarize(avg = mean(.epred))

h5c_amces <- h5c_mms %>% 
  group_by(feat_govt, feat_funding_short) %>% 
  compare_levels(variable = avg, by = feat_transp_short, comparison = "control")

Code

p_h5c_mms <- h5c_mms %>%
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_percent()) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  facet_nested_wrap(
    vars(feat_govt, feat_funding_short),
    ncol = 3,
    strip = strip_nested(
      text_x = list(element_text(
        family = "Libre Franklin",
        face = "bold"
      ), NULL),
      background_x = list(element_rect(fill = "grey96"), NULL),
      by_layer_x = TRUE
    )
  ) +
  labs(
    x = "Overall average predicted probability",
    y = NULL,
    fill = NULL,
    title = "Estimated marginal means"
  ) +
  theme(
    legend.position = "top",
    legend.justification = "left",
    legend.margin = margin(l = -7, t = 0)
  )
p_h5c_mms

Code


baselines <- expand_grid(
  avg = 0,
  feat_transp_short = "No",
  feat_govt = levels(h5c_mms$feat_govt),
  feat_funding_short = levels(h5c_mms$feat_funding_short)
)

p_h5c_amces <- h5c_amces %>% 
  ungroup() %>% 
  separate_wider_delim(
    feat_transp_short,
    delim = " - ", 
    names = c("feat_transp_short", "reference_level")
  ) %>% 
  bind_rows(baselines) %>% 
  mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5c_mms$feat_transp_short))) %>% 
  mutate(feat_govt = factor(feat_govt, levels = levels(h5c_mms$feat_govt))) %>% 
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_pp) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  facet_nested_wrap(
    vars(feat_govt, feat_funding_short),
    ncol = 3,
    strip = strip_nested(
      text_x = list(element_text(
        family = "Libre Franklin",
        face = "bold"
      ), NULL),
      background_x = list(element_rect(fill = "grey96"), NULL),
      by_layer_x = TRUE
    )
  ) +
  labs(
    x = "Percentage point change in probability of choice selection",
    y = NULL,
    fill = NULL,
    title = "Difference in estimated marginal means"
  )

p_h5c_amces

H_5d: Relationship with host government, organizational practices, and issue area

Hypothesis 5_d: Relationship with host government, organizational practices, and issue area

Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are also financially transparent and work in less contentious areas (emergency response and refugee relief)

Estimand

TODO

Code

h5d_mms <- preds_all %>% 
  group_by(feat_transp_short, feat_govt, feat_issue, .draw) %>% 
  summarize(avg = mean(.epred))

h5d_amces <- h5d_mms %>% 
  group_by(feat_govt, feat_issue) %>% 
  compare_levels(variable = avg, by = feat_transp_short, comparison = "control")

Code

p_h5d_mms <- h5d_mms %>%
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_percent()) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  facet_nested_wrap(
    vars(feat_govt, feat_issue),
    ncol = 4,
    strip = strip_nested(
      text_x = list(element_text(
        family = "Libre Franklin",
        face = "bold"
      ), NULL),
      background_x = list(element_rect(fill = "grey96"), NULL),
      by_layer_x = TRUE
    )
  ) +
  labs(
    x = "Overall average predicted probability",
    y = NULL,
    fill = NULL,
    title = "Estimated marginal means"
  ) +
  theme(
    legend.position = "top",
    legend.justification = "left",
    legend.margin = margin(l = -7, t = 0)
  )
p_h5d_mms

Code


baselines <- expand_grid(
  avg = 0,
  feat_transp_short = "No",
  feat_govt = levels(h5d_mms$feat_govt),
  feat_issue = levels(h5d_mms$feat_issue)
)

p_h5d_amces <- h5d_amces %>% 
  ungroup() %>% 
  separate_wider_delim(
    feat_transp_short,
    delim = " - ", 
    names = c("feat_transp_short", "reference_level")
  ) %>% 
  bind_rows(baselines) %>% 
  mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5d_mms$feat_transp_short))) %>% 
  mutate(feat_govt = factor(feat_govt, levels = levels(h5d_mms$feat_govt))) %>% 
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_pp) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  facet_nested_wrap(
    vars(feat_govt, feat_issue),
    ncol = 4,
    strip = strip_nested(
      text_x = list(element_text(
        family = "Libre Franklin",
        face = "bold"
      ), NULL),
      background_x = list(element_rect(fill = "grey96"), NULL),
      by_layer_x = TRUE
    )
  ) +
  labs(
    x = "Percentage point change in probability of choice selection",
    y = NULL,
    fill = NULL,
    title = "Difference in estimated marginal means"
  )

p_h5d_amces

H_5e: Relationship with host government, organizational practices, issue area, and funding sources

Hypothesis 5_e: Relationship with host government, organizational practices, issue area, and funding sources

TODO

Code

h5e_mms <- preds_all %>% 
  group_by(feat_transp_short, feat_govt, feat_funding_short, feat_issue, .draw) %>% 
  summarize(avg = mean(.epred))

h5e_amces <- h5e_mms %>% 
  group_by(feat_govt, feat_funding_short, feat_issue) %>% 
  compare_levels(variable = avg, by = feat_transp_short, comparison = "control")

Code

p_h5e_mms <- h5e_mms %>%
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_percent()) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  facet_nested_wrap(
    vars(feat_govt, feat_funding_short, feat_issue),
    ncol = 12,
    strip = strip_nested(
      text_x = list(element_text(
        family = "Libre Franklin",
        face = "bold"
      ), NULL),
      background_x = list(element_rect(fill = "grey96"), NULL),
      by_layer_x = TRUE
    )
  ) +
  labs(
    x = "Overall average predicted probability",
    y = NULL,
    fill = NULL,
    title = "Estimated marginal means"
  ) +
  theme(
    legend.position = "top",
    legend.justification = "left",
    legend.margin = margin(l = -7, t = 0)
  )
p_h5e_mms

Code


baselines <- expand_grid(
  avg = 0,
  feat_transp_short = "No",
  feat_govt = levels(h5e_mms$feat_govt),
  feat_funding_short = levels(h5e_mms$feat_funding_short),
  feat_issue = levels(h5e_mms$feat_issue)
)

p_h5e_amces <- h5e_amces %>% 
  ungroup() %>% 
  separate_wider_delim(
    feat_transp_short,
    delim = " - ", 
    names = c("feat_transp_short", "reference_level")
  ) %>% 
  bind_rows(baselines) %>% 
  mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5e_mms$feat_transp_short))) %>% 
  mutate(feat_govt = factor(feat_govt, levels = levels(h5e_mms$feat_govt))) %>% 
  ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_pp) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") +
  facet_nested_wrap(
    vars(feat_govt, feat_funding_short, feat_issue),
    ncol = 12,
    strip = strip_nested(
      text_x = list(element_text(
        family = "Libre Franklin",
        face = "bold"
      ), NULL),
      background_x = list(element_rect(fill = "grey96"), NULL),
      by_layer_x = TRUE
    )
  ) +
  labs(
    x = "Percentage point change in probability of choice selection",
    y = NULL,
    fill = NULL,
    title = "Difference in estimated marginal means"
  )

p_h5e_amces

H_5f: Accountability

Hypothesis 5_f: Accountability

Donors should be no more or less likely to donate to NGOs that are accountable and hold regular third party audits

Mechanism: Donors don’t necessarily seek assurance through third-party programs/audits and charity watchdogs, but rather through word of mouth, personal scrutiny and local networks

Estimand

\[ \begin{aligned} \theta =&\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Accountability} = \text{Yes} \right) \bigr] \end{aligned} \]

Code

h5f_mms <- preds_all %>% 
  group_by(feat_acc_short, .draw) %>% 
  summarize(avg = mean(.epred))

h5f_amces <- h5f_mms %>% 
  group_by(feat_acc_short) %>% 
  compare_levels(variable = avg, by = feat_acc_short, comparison = "control")

Code

p_h5f_mms <- h5f_mms %>% 
  ggplot(aes(x = avg, y = feat_acc_short, fill = feat_acc_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_percent()) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(9, 4)], guide = "none") +
  labs(
    x = "Overall average predicted probability",
    y = NULL,
    fill = NULL,
    title = "Estimated marginal means"
  )

p_h5f_amces <- h5f_amces %>% 
  ungroup() %>% 
  separate_wider_delim(
    feat_acc_short,
    delim = " - ", 
    names = c("feat_acc_short", "reference_level")
  ) %>% 
  add_row(avg = 0, feat_acc_short = unique(.$reference_level)) %>%
  mutate(feat_acc_short = factor(feat_acc_short, levels = levels(h5f_mms$feat_acc_short))) %>% 
  ggplot(aes(x = avg, y = feat_acc_short, fill = feat_acc_short)) +
  stat_halfeye() +
  geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) +
  scale_x_continuous(labels = label_pp) +
  scale_y_discrete(labels = label_wrap(11)) +
  scale_fill_manual(values = clrs$prism[c(9, 4)], guide = "none") +
  labs(
    x = "Percentage point change in probability of choice selection",
    y = NULL,
    fill = NULL,
    title = "Average marginal component effects (AMCEs)"
  )

p_h5f_mms | p_h5f_amces

Code

h5f_tbl_mm <- h5f_mms %>% 
  group_by(feat_acc_short) %>% 
  median_qi() %>% 
  mutate(nice = glue("{fmt_decimal(avg)}<br>{build_ci(.lower, .upper)}")) %>% 
  arrange(desc(feat_acc_short)) %>% 
  select(feat_acc_short, nice)

h5f_tbl_amces <- h5f_amces %>% 
  group_by(feat_acc_short) %>% 
  summarize(
    median_qi(avg),
    p_gt_0 = sum(avg > 0) / n()
  ) %>%
  mutate(p_neq_0 = ifelse(y >= 0, p_gt_0, 1 - p_gt_0)) %>% 
  mutate(nice = glue("{fmt_decimal(y)}<br>{build_ci(ymin, ymax)}")) %>% 
  mutate(feat_acc_short = str_replace(feat_acc_short, " - ", "−<br>")) %>% 
  mutate(across(starts_with("p_"), ~fmt_decimal(.))) %>% 
  arrange(desc(feat_acc_short)) %>% 
  select(contrast = feat_acc_short, amce_nice = nice, p_neq_0)

bind_cols(
  h5f_tbl_mm, 
  add_row(h5f_tbl_amces, contrast = "*(Reference)*")
) %>% 
  gt() %>% 
  sub_missing(columns = everything(), missing_text = "—") %>% 
  fmt_markdown(columns = c(nice, amce_nice, contrast)) %>%
  cols_align(align = "center", columns = everything()) %>% 
  cols_align(align = "left", columns = c(feat_acc_short, contrast)) %>% 
  cols_label(
    feat_acc_short = "Accountability", 
    nice = "Posterior EMM",
    contrast = "Contrast",
    amce_nice = "Posterior AMCE",
    p_neq_0 = "*p*~direction~",
    .fn = md
  ) %>% 
  tab_style(
    style = cell_text(v_align = "top"),
    locations = cells_body()
  ) %>% 
  tab_footnote(
    footnote = "Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.",
    locations = cells_column_labels(columns = c(nice, amce_nice))
  ) %>% 
  tab_footnote(
    footnote = md("The probability of direction (*p*~direction~) is the probability that the posterior AMCE is strictly positive or negative—it is the proportion of the posterior AMCE that is the sign of the median."),
    locations = cells_column_labels(columns = p_neq_0)
  ) %>% 
  gt_add_divider(columns = nice, style = "dashed", weight = px(1)) %>% 
  opt_footnote_marks(marks = "standard") %>% 
  opt_horizontal_padding(3) %>% 
  opts_theme()

Accountability	Posterior EMM^*	Contrast	Posterior AMCE^*	p_direction^†
Yes	0.306 [0.299, 0.313]	Yes− No	0.101 [0.092, 0.110]	1.000
No	0.205 [0.198, 0.211]	(Reference)	—	—
^* Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.
^† The probability of direction (p_direction) is the probability that the posterior AMCE is strictly positive or negative—it is the proportion of the posterior AMCE that is the sign of the median.

--- title: "H~5~: Organizational practices" format: html: code-fold: true --- ```{r setup, include=FALSE} knitr::opts_chunk$set( fig.align = "center", fig.retina = 3, fig.width = 6, fig.height = (6 * 0.618), out.width = "80%", collapse = TRUE, dev = "ragg_png" ) options( digits = 3, width = 120, dplyr.summarise.inform = FALSE, knitr.kable.NA = "" ) ``` ```{r libraries-data, warning=FALSE, message=FALSE} library(tidyverse) library(targets) library(tidybayes) library(patchwork) library(scales) library(ggh4x) library(glue) library(gt) library(gtExtras) tar_config_set( store = here::here("_targets"), script = here::here("_targets.R") ) preds_all <- tar_read(preds_conditional_treatment_only) invisible(list2env(tar_read(graphic_functions), .GlobalEnv)) invisible(list2env(tar_read(table_functions), .GlobalEnv)) theme_set(theme_ngo()) ``` # H~5a~: Financial transparency ::: {.callout-tip icon=false} ## {{< iconify fa6-solid flask-vial >}} Hypothesis 5~a~: Financial transparency **Donors will show increased willingness to donate to NGOs that are financially transparent** *Mechanism: Perception of efficacy* ::: ## Estimand $$ \begin{aligned} \theta =&\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Transparency} = \text{Yes} \right) \bigr] \end{aligned} $$ ```{r h5a-calc-estimands} h5a_mms <- preds_all %>% group_by(feat_transp_short, .draw) %>% summarize(avg = mean(.epred)) h5a_amces <- h5a_mms %>% group_by(feat_transp_short) %>% compare_levels(variable = avg, by = feat_transp_short, comparison = "control") ``` ```{r plot-h5a-mm-amce, fig.width=10, fig.height=4, out.width="100%"} #| column: body-outset-right p_h5a_mms <- h5a_mms %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_percent()) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + labs( x = "Overall average predicted probability", y = NULL, fill = NULL, title = "Estimated marginal means" ) p_h5a_amces <- h5a_amces %>% ungroup() %>% separate_wider_delim( feat_transp_short, delim = " - ", names = c("feat_transp_short", "reference_level") ) %>% add_row(avg = 0, feat_transp_short = unique(.$reference_level)) %>% mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5a_mms$feat_transp_short))) %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_pp) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + labs( x = "Percentage point change in probability of choice selection", y = NULL, fill = NULL, title = "Average marginal component effects (AMCEs)" ) p_h5a_mms | p_h5a_amces ``` ```{r table-h5a-mm-amce} #| column: body-outset-right h5a_tbl_mm <- h5a_mms %>% group_by(feat_transp_short) %>% median_qi() %>% mutate(nice = glue("{fmt_decimal(avg)} {build_ci(.lower, .upper)}")) %>% arrange(desc(feat_transp_short)) %>% select(feat_transp_short, nice) h5a_tbl_amces <- h5a_amces %>% group_by(feat_transp_short) %>% summarize( median_qi(avg), p_gt_0 = sum(avg > 0) / n() ) %>% mutate(p_neq_0 = ifelse(y >= 0, p_gt_0, 1 - p_gt_0)) %>% mutate(nice = glue("{fmt_decimal(y)} {build_ci(ymin, ymax)}")) %>% mutate(feat_transp_short = str_replace(feat_transp_short, " - ", "− ")) %>% mutate(across(starts_with("p_"), ~fmt_decimal(.))) %>% arrange(desc(feat_transp_short)) %>% select(contrast = feat_transp_short, amce_nice = nice, p_neq_0) bind_cols( h5a_tbl_mm, add_row(h5a_tbl_amces, contrast = "*(Reference)*") ) %>% gt() %>% sub_missing(columns = everything(), missing_text = "—") %>% fmt_markdown(columns = c(nice, amce_nice, contrast)) %>% cols_align(align = "center", columns = everything()) %>% cols_align(align = "left", columns = c(feat_transp_short, contrast)) %>% cols_label( feat_transp_short = "Transparency", nice = "Posterior EMM", contrast = "Contrast", amce_nice = "Posterior AMCE", p_neq_0 = "*p*~direction~", .fn = md ) %>% tab_style( style = cell_text(v_align = "top"), locations = cells_body() ) %>% tab_footnote( footnote = "Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.", locations = cells_column_labels(columns = c(nice, amce_nice)) ) %>% tab_footnote( footnote = md("The probability of direction (*p*~direction~) is the probability that the posterior AMCE is strictly positive or negative—it is the proportion of the posterior AMCE that is the sign of the median."), locations = cells_column_labels(columns = p_neq_0) ) %>% gt_add_divider(columns = nice, style = "dashed", weight = px(1)) %>% opt_footnote_marks(marks = "standard") %>% opt_horizontal_padding(3) %>% opts_theme() ``` # H~5b~: Relationship with host government and organizational practices ## Transparency ::: {.callout-tip icon=false} ### {{< iconify fa6-solid flask-vial >}} Hypothesis 5~b~: Relationship with host government, organizational practices **Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are also financially transparent** ::: $$ \begin{aligned} \theta =&\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Transparency} = \text{Yes} \mid \text{Relationship = Under crackdown or Criticized} \right) \bigr] - \\ &\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Transparency} = \text{Yes} \mid \text{Relationship = Friendly} \right) \bigr] \end{aligned} $$ ```{r h5b-calc-estimands} h5b_mms <- preds_all %>% group_by(feat_transp_short, feat_govt, .draw) %>% summarize(avg = mean(.epred)) h5b_amces <- h5b_mms %>% group_by(feat_govt) %>% compare_levels(variable = avg, by = feat_transp_short, comparison = "control") ``` ```{r plot-h5b-mm-amce, fig.width=10, fig.height=6.5, out.width="100%"} #| column: body-outset-right p_h5b_mms <- h5b_mms %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_percent()) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + facet_wrap(vars(feat_govt), ncol = 1) + labs( x = "Overall average predicted probability", y = NULL, fill = NULL, title = "Estimated marginal means" ) + theme( legend.position = "top", legend.justification = "left", legend.margin = margin(l = -7, t = 0) ) p_h5b_amces <- h5b_amces %>% ungroup() %>% separate_wider_delim( feat_transp_short, delim = " - ", names = c("feat_transp_short", "reference_level") ) %>% add_row(avg = 0, feat_transp_short = "No", feat_govt = "Friendly relationship with government") %>% add_row(avg = 0, feat_transp_short = "No", feat_govt = "Criticized by government") %>% add_row(avg = 0, feat_transp_short = "No", feat_govt = "Under government crackdown") %>% mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5b_mms$feat_transp_short))) %>% mutate(feat_govt = factor(feat_govt, levels = levels(h5b_mms$feat_govt))) %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_pp) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + facet_wrap(vars(feat_govt), ncol = 1) + labs( x = "Percentage point change in probability of choice selection", y = NULL, fill = NULL, title = "Difference in estimated marginal means" ) p_h5b_mms | p_h5b_amces ``` ```{r table-h5b-mm-amce} #| column: body-outset-right h5b_tbl_mm <- h5b_mms %>% group_by(feat_govt, feat_transp_short) %>% median_qi() %>% mutate(nice = glue("{fmt_decimal(avg)} {build_ci(.lower, .upper)}")) %>% arrange(desc(feat_govt), desc(feat_transp_short)) %>% select(feat_transp_short, feat_govt, nice) h5b_tbl_amces <- h5b_amces %>% group_by(feat_govt, feat_transp_short) %>% summarize( median_qi(avg), p_gt_0 = sum(avg > 0) / n() ) %>% ungroup() %>% mutate(p_neq_0 = ifelse(y >= 0, p_gt_0, 1 - p_gt_0)) %>% mutate(nice = glue("{fmt_decimal(y)} {build_ci(ymin, ymax)}")) %>% mutate(feat_transp_short = str_replace(feat_transp_short, " - ", "− ")) %>% mutate(across(starts_with("p_"), ~fmt_decimal(.))) %>% arrange(desc(feat_govt)) %>% select(contrast = feat_transp_short, diff_nice = nice, p_neq_0) bind_cols( h5b_tbl_mm, h5b_tbl_amces %>% add_row(contrast = NA, .after = 1) %>% add_row(contrast = NA, .after = 3) %>% add_row(contrast = NA, .after = 5) ) %>% mutate(feat_govt = fct_relabel(feat_govt, ~paste("Relationship with government:", .x))) %>% group_by(feat_govt) %>% gt() %>% sub_missing(columns = everything(), missing_text = "—") %>% fmt_markdown(columns = c(nice, diff_nice, contrast)) %>% cols_align(align = "center", columns = everything()) %>% cols_align(align = "left", columns = c(feat_transp_short, contrast)) %>% cols_label( feat_transp_short = "Transparency", nice = "Posterior EMM", contrast = "Contrast", diff_nice = "Posterior ∆", p_neq_0 = "*p*~direction~", .fn = md ) %>% tab_style( style = cell_text(v_align = "top"), locations = cells_body() ) %>% tab_style( style = cell_fill(color = "grey90"), locations = cells_row_groups() ) %>% tab_footnote( footnote = "Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.", locations = cells_column_labels(columns = c(nice, diff_nice)) ) %>% tab_footnote( footnote = md("The probability of direction (*p*~direction~) is the probability that the posterior difference in marginal means is strictly positive or negative—it is the proportion of the posterior difference in marginal means that is the sign of the median."), locations = cells_column_labels(columns = p_neq_0) ) %>% gt_add_divider(columns = nice, style = "dashed", weight = px(1)) %>% opt_footnote_marks(marks = "standard") %>% opt_horizontal_padding(3) %>% opts_theme() ``` ## Accountability ::: {.callout-tip icon=false} ### {{< iconify fa6-solid flask-vial >}} Hypothesis 5~b2~: Relationship with host government, organizational practices **Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are accountable and hold regular third party audits** ::: $$ \begin{aligned} \theta =&\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Accountability} = \text{Yes} \mid \text{Relationship = Under crackdown or Criticized} \right) \bigr] - \\ &\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Accountability} = \text{Yes} \mid \text{Relationship = Friendly} \right) \bigr] \end{aligned} $$ ```{r h5b2-calc-estimands} h5b2_mms <- preds_all %>% group_by(feat_acc_short, feat_govt, .draw) %>% summarize(avg = mean(.epred)) h5b2_amces <- h5b2_mms %>% group_by(feat_govt) %>% compare_levels(variable = avg, by = feat_acc_short, comparison = "control") ``` ```{r plot-h5b2-mm-amce, fig.width=10, fig.height=6.5, out.width="100%"} #| column: body-outset-right p_h5b2_mms <- h5b2_mms %>% ggplot(aes(x = avg, y = feat_acc_short, fill = feat_acc_short)) + stat_halfeye() + geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_percent()) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(9, 4)], guide = "none") + facet_wrap(vars(feat_govt), ncol = 1) + labs( x = "Overall average predicted probability", y = NULL, fill = NULL, title = "Estimated marginal means" ) + theme( legend.position = "top", legend.justification = "left", legend.margin = margin(l = -7, t = 0) ) p_h5b2_amces <- h5b2_amces %>% ungroup() %>% separate_wider_delim( feat_acc_short, delim = " - ", names = c("feat_acc_short", "reference_level") ) %>% add_row(avg = 0, feat_acc_short = "No", feat_govt = "Friendly relationship with government") %>% add_row(avg = 0, feat_acc_short = "No", feat_govt = "Criticized by government") %>% add_row(avg = 0, feat_acc_short = "No", feat_govt = "Under government crackdown") %>% mutate(feat_acc_short = factor(feat_acc_short, levels = levels(h5b2_mms$feat_acc_short))) %>% mutate(feat_govt = factor(feat_govt, levels = levels(h5b2_mms$feat_govt))) %>% ggplot(aes(x = avg, y = feat_acc_short, fill = feat_acc_short)) + stat_halfeye() + geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_pp) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(9, 4)], guide = "none") + facet_wrap(vars(feat_govt), ncol = 1) + labs( x = "Percentage point change in probability of choice selection", y = NULL, fill = NULL, title = "Difference in estimated marginal means" ) p_h5b2_mms | p_h5b2_amces ``` ```{r table-h5b2-mm-amce} #| column: body-outset-right h5b2_tbl_mm <- h5b2_mms %>% group_by(feat_govt, feat_acc_short) %>% median_qi() %>% mutate(nice = glue("{fmt_decimal(avg)} {build_ci(.lower, .upper)}")) %>% arrange(desc(feat_govt), desc(feat_acc_short)) %>% select(feat_acc_short, feat_govt, nice) h5b2_tbl_amces <- h5b2_amces %>% group_by(feat_govt, feat_acc_short) %>% summarize( median_qi(avg), p_gt_0 = sum(avg > 0) / n() ) %>% ungroup() %>% mutate(p_neq_0 = ifelse(y >= 0, p_gt_0, 1 - p_gt_0)) %>% mutate(nice = glue("{fmt_decimal(y)} {build_ci(ymin, ymax)}")) %>% mutate(feat_acc_short = str_replace(feat_acc_short, " - ", "− ")) %>% mutate(across(starts_with("p_"), ~fmt_decimal(.))) %>% arrange(desc(feat_govt)) %>% select(contrast = feat_acc_short, diff_nice = nice, p_neq_0) bind_cols( h5b2_tbl_mm, h5b2_tbl_amces %>% add_row(contrast = NA, .after = 1) %>% add_row(contrast = NA, .after = 3) %>% add_row(contrast = NA, .after = 5) ) %>% mutate(feat_govt = fct_relabel(feat_govt, ~paste("Relationship with government:", .x))) %>% group_by(feat_govt) %>% gt() %>% sub_missing(columns = everything(), missing_text = "—") %>% fmt_markdown(columns = c(nice, diff_nice, contrast)) %>% cols_align(align = "center", columns = everything()) %>% cols_align(align = "left", columns = c(feat_acc_short, contrast)) %>% cols_label( feat_acc_short = "Accountability", nice = "Posterior EMM", contrast = "Contrast", diff_nice = "Posterior ∆", p_neq_0 = "*p*~direction~", .fn = md ) %>% tab_style( style = cell_text(v_align = "top"), locations = cells_body() ) %>% tab_style( style = cell_fill(color = "grey90"), locations = cells_row_groups() ) %>% tab_footnote( footnote = "Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.", locations = cells_column_labels(columns = c(nice, diff_nice)) ) %>% tab_footnote( footnote = md("The probability of direction (*p*~direction~) is the probability that the posterior difference in marginal means is strictly positive or negative—it is the proportion of the posterior difference in marginal means that is the sign of the median."), locations = cells_column_labels(columns = p_neq_0) ) %>% gt_add_divider(columns = nice, style = "dashed", weight = px(1)) %>% opt_footnote_marks(marks = "standard") %>% opt_horizontal_padding(3) %>% opts_theme() ``` ## H~5c~: Relationship with host government, organizational practices, and funding sources ::: {.callout-tip icon=false} ## {{< iconify fa6-solid flask-vial >}} Hypothesis 5~c~: Relationship with host government, organizational practices, and funding sources **Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are also financially transparent and are funded primarily by numerous small private donors** ::: ## Estimand TODO ```{r h5c-calc-estimands} h5c_mms <- preds_all %>% group_by(feat_transp_short, feat_govt, feat_funding_short, .draw) %>% summarize(avg = mean(.epred)) h5c_amces <- h5c_mms %>% group_by(feat_govt, feat_funding_short) %>% compare_levels(variable = avg, by = feat_transp_short, comparison = "control") ``` ```{r plot-h5c-mm-amce, fig.width=10, fig.height=6.5, out.width="100%"} #| column: body-outset-right p_h5c_mms <- h5c_mms %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_percent()) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + facet_nested_wrap( vars(feat_govt, feat_funding_short), ncol = 3, strip = strip_nested( text_x = list(element_text( family = "Libre Franklin", face = "bold" ), NULL), background_x = list(element_rect(fill = "grey96"), NULL), by_layer_x = TRUE ) ) + labs( x = "Overall average predicted probability", y = NULL, fill = NULL, title = "Estimated marginal means" ) + theme( legend.position = "top", legend.justification = "left", legend.margin = margin(l = -7, t = 0) ) p_h5c_mms baselines <- expand_grid( avg = 0, feat_transp_short = "No", feat_govt = levels(h5c_mms$feat_govt), feat_funding_short = levels(h5c_mms$feat_funding_short) ) p_h5c_amces <- h5c_amces %>% ungroup() %>% separate_wider_delim( feat_transp_short, delim = " - ", names = c("feat_transp_short", "reference_level") ) %>% bind_rows(baselines) %>% mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5c_mms$feat_transp_short))) %>% mutate(feat_govt = factor(feat_govt, levels = levels(h5c_mms$feat_govt))) %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_pp) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + facet_nested_wrap( vars(feat_govt, feat_funding_short), ncol = 3, strip = strip_nested( text_x = list(element_text( family = "Libre Franklin", face = "bold" ), NULL), background_x = list(element_rect(fill = "grey96"), NULL), by_layer_x = TRUE ) ) + labs( x = "Percentage point change in probability of choice selection", y = NULL, fill = NULL, title = "Difference in estimated marginal means" ) p_h5c_amces ``` # H~5d~: Relationship with host government, organizational practices, and issue area ::: {.callout-tip icon=false} ## {{< iconify fa6-solid flask-vial >}} Hypothesis 5~d~: Relationship with host government, organizational practices, and issue area **Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are also financially transparent and work in less contentious areas (emergency response and refugee relief)** ::: ## Estimand TODO ```{r h5d-calc-estimands} h5d_mms <- preds_all %>% group_by(feat_transp_short, feat_govt, feat_issue, .draw) %>% summarize(avg = mean(.epred)) h5d_amces <- h5d_mms %>% group_by(feat_govt, feat_issue) %>% compare_levels(variable = avg, by = feat_transp_short, comparison = "control") ``` ```{r plot-h5d-mm-amce, fig.width=10, fig.height=6.5, out.width="100%"} #| column: body-outset-right p_h5d_mms <- h5d_mms %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_percent()) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + facet_nested_wrap( vars(feat_govt, feat_issue), ncol = 4, strip = strip_nested( text_x = list(element_text( family = "Libre Franklin", face = "bold" ), NULL), background_x = list(element_rect(fill = "grey96"), NULL), by_layer_x = TRUE ) ) + labs( x = "Overall average predicted probability", y = NULL, fill = NULL, title = "Estimated marginal means" ) + theme( legend.position = "top", legend.justification = "left", legend.margin = margin(l = -7, t = 0) ) p_h5d_mms baselines <- expand_grid( avg = 0, feat_transp_short = "No", feat_govt = levels(h5d_mms$feat_govt), feat_issue = levels(h5d_mms$feat_issue) ) p_h5d_amces <- h5d_amces %>% ungroup() %>% separate_wider_delim( feat_transp_short, delim = " - ", names = c("feat_transp_short", "reference_level") ) %>% bind_rows(baselines) %>% mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5d_mms$feat_transp_short))) %>% mutate(feat_govt = factor(feat_govt, levels = levels(h5d_mms$feat_govt))) %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_pp) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + facet_nested_wrap( vars(feat_govt, feat_issue), ncol = 4, strip = strip_nested( text_x = list(element_text( family = "Libre Franklin", face = "bold" ), NULL), background_x = list(element_rect(fill = "grey96"), NULL), by_layer_x = TRUE ) ) + labs( x = "Percentage point change in probability of choice selection", y = NULL, fill = NULL, title = "Difference in estimated marginal means" ) p_h5d_amces ``` # H~5e~: Relationship with host government, organizational practices, issue area, and funding sources ::: {.callout-tip icon=false} ## {{< iconify fa6-solid flask-vial >}} Hypothesis 5~e~: Relationship with host government, organizational practices, issue area, and funding sources **Donors will show increased willingness to donate to NGOs that are criticized by the government/under government crackdown when they are also financially transparent and work in less contentious areas (emergency response and refugee relief) and are funded by numerous small donors** ::: TODO ```{r h5e-calc-estimands} h5e_mms <- preds_all %>% group_by(feat_transp_short, feat_govt, feat_funding_short, feat_issue, .draw) %>% summarize(avg = mean(.epred)) h5e_amces <- h5e_mms %>% group_by(feat_govt, feat_funding_short, feat_issue) %>% compare_levels(variable = avg, by = feat_transp_short, comparison = "control") ``` ```{r plot-h5e-mm-amce, fig.width=14, fig.height=6.5, out.width="100%"} #| column: page p_h5e_mms <- h5e_mms %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_percent()) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + facet_nested_wrap( vars(feat_govt, feat_funding_short, feat_issue), ncol = 12, strip = strip_nested( text_x = list(element_text( family = "Libre Franklin", face = "bold" ), NULL), background_x = list(element_rect(fill = "grey96"), NULL), by_layer_x = TRUE ) ) + labs( x = "Overall average predicted probability", y = NULL, fill = NULL, title = "Estimated marginal means" ) + theme( legend.position = "top", legend.justification = "left", legend.margin = margin(l = -7, t = 0) ) p_h5e_mms baselines <- expand_grid( avg = 0, feat_transp_short = "No", feat_govt = levels(h5e_mms$feat_govt), feat_funding_short = levels(h5e_mms$feat_funding_short), feat_issue = levels(h5e_mms$feat_issue) ) p_h5e_amces <- h5e_amces %>% ungroup() %>% separate_wider_delim( feat_transp_short, delim = " - ", names = c("feat_transp_short", "reference_level") ) %>% bind_rows(baselines) %>% mutate(feat_transp_short = factor(feat_transp_short, levels = levels(h5e_mms$feat_transp_short))) %>% mutate(feat_govt = factor(feat_govt, levels = levels(h5e_mms$feat_govt))) %>% ggplot(aes(x = avg, y = feat_transp_short, fill = feat_transp_short)) + stat_halfeye() + geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_pp) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(10, 3)], guide = "none") + facet_nested_wrap( vars(feat_govt, feat_funding_short, feat_issue), ncol = 12, strip = strip_nested( text_x = list(element_text( family = "Libre Franklin", face = "bold" ), NULL), background_x = list(element_rect(fill = "grey96"), NULL), by_layer_x = TRUE ) ) + labs( x = "Percentage point change in probability of choice selection", y = NULL, fill = NULL, title = "Difference in estimated marginal means" ) p_h5e_amces ``` # H~5f~: Accountability ::: {.callout-tip icon=false} ## {{< iconify fa6-solid flask-vial >}} Hypothesis 5~f~: Accountability **Donors should be no more or less likely to donate to NGOs that are accountable and hold regular third party audits** *Mechanism: Donors don't necessarily seek assurance through third-party programs/audits and charity watchdogs, but rather through word of mouth, personal scrutiny and local networks* ::: ## Estimand $$ \begin{aligned} \theta =&\ \textbf{E}\bigl[ Y_i \mid \operatorname{do}\left( \text{Accountability} = \text{Yes} \right) \bigr] \end{aligned} $$ ```{r h5f-calc-estimands} h5f_mms <- preds_all %>% group_by(feat_acc_short, .draw) %>% summarize(avg = mean(.epred)) h5f_amces <- h5f_mms %>% group_by(feat_acc_short) %>% compare_levels(variable = avg, by = feat_acc_short, comparison = "control") ``` ```{r plot-h5f-mm-amce, fig.width=10, fig.height=4, out.width="100%"} #| column: body-outset-right p_h5f_mms <- h5f_mms %>% ggplot(aes(x = avg, y = feat_acc_short, fill = feat_acc_short)) + stat_halfeye() + geom_vline(xintercept = 0.25, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_percent()) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(9, 4)], guide = "none") + labs( x = "Overall average predicted probability", y = NULL, fill = NULL, title = "Estimated marginal means" ) p_h5f_amces <- h5f_amces %>% ungroup() %>% separate_wider_delim( feat_acc_short, delim = " - ", names = c("feat_acc_short", "reference_level") ) %>% add_row(avg = 0, feat_acc_short = unique(.$reference_level)) %>% mutate(feat_acc_short = factor(feat_acc_short, levels = levels(h5f_mms$feat_acc_short))) %>% ggplot(aes(x = avg, y = feat_acc_short, fill = feat_acc_short)) + stat_halfeye() + geom_vline(xintercept = 0, color = clrs$prism[8], linetype = "dashed", linewidth = 0.25) + scale_x_continuous(labels = label_pp) + scale_y_discrete(labels = label_wrap(11)) + scale_fill_manual(values = clrs$prism[c(9, 4)], guide = "none") + labs( x = "Percentage point change in probability of choice selection", y = NULL, fill = NULL, title = "Average marginal component effects (AMCEs)" ) p_h5f_mms | p_h5f_amces ``` ```{r table-h5f-mm-amce} #| column: body-outset-right h5f_tbl_mm <- h5f_mms %>% group_by(feat_acc_short) %>% median_qi() %>% mutate(nice = glue("{fmt_decimal(avg)} {build_ci(.lower, .upper)}")) %>% arrange(desc(feat_acc_short)) %>% select(feat_acc_short, nice) h5f_tbl_amces <- h5f_amces %>% group_by(feat_acc_short) %>% summarize( median_qi(avg), p_gt_0 = sum(avg > 0) / n() ) %>% mutate(p_neq_0 = ifelse(y >= 0, p_gt_0, 1 - p_gt_0)) %>% mutate(nice = glue("{fmt_decimal(y)} {build_ci(ymin, ymax)}")) %>% mutate(feat_acc_short = str_replace(feat_acc_short, " - ", "− ")) %>% mutate(across(starts_with("p_"), ~fmt_decimal(.))) %>% arrange(desc(feat_acc_short)) %>% select(contrast = feat_acc_short, amce_nice = nice, p_neq_0) bind_cols( h5f_tbl_mm, add_row(h5f_tbl_amces, contrast = "*(Reference)*") ) %>% gt() %>% sub_missing(columns = everything(), missing_text = "—") %>% fmt_markdown(columns = c(nice, amce_nice, contrast)) %>% cols_align(align = "center", columns = everything()) %>% cols_align(align = "left", columns = c(feat_acc_short, contrast)) %>% cols_label( feat_acc_short = "Accountability", nice = "Posterior EMM", contrast = "Contrast", amce_nice = "Posterior AMCE", p_neq_0 = "*p*~direction~", .fn = md ) %>% tab_style( style = cell_text(v_align = "top"), locations = cells_body() ) %>% tab_footnote( footnote = "Values are on the percentage-point scale; single value is posterior median; 95% credible interval in brackets.", locations = cells_column_labels(columns = c(nice, amce_nice)) ) %>% tab_footnote( footnote = md("The probability of direction (*p*~direction~) is the probability that the posterior AMCE is strictly positive or negative—it is the proportion of the posterior AMCE that is the sign of the median."), locations = cells_column_labels(columns = p_neq_0) ) %>% gt_add_divider(columns = nice, style = "dashed", weight = px(1)) %>% opt_footnote_marks(marks = "standard") %>% opt_horizontal_padding(3) %>% opts_theme() ```

H5a: Financial transparency

Estimand

H5b: Relationship with host government and organizational practices

Transparency

Accountability

H5c: Relationship with host government, organizational practices, and funding sources

Estimand

H5d: Relationship with host government, organizational practices, and issue area

Estimand

H5e: Relationship with host government, organizational practices, issue area, and funding sources

H5f: Accountability

Estimand

H_5a: Financial transparency

H_5b: Relationship with host government and organizational practices

H_5c: Relationship with host government, organizational practices, and funding sources

H_5d: Relationship with host government, organizational practices, and issue area

H_5e: Relationship with host government, organizational practices, issue area, and funding sources

H_5f: Accountability