I am running an RCT in 80 government and 40 low-cost private schools in Cape Coast Ghana for first grade students. These schools get one or both of two cross-randomized treatments. One treatment organizes scripted parent meetings where parents are 1) told the benefits of doing literacy activities with their child at home (e.g. asking what they learned in school, reading with them, going through school materials etc.), 2) trained on how to understand the report card they get on students literacy scores, and 3) briefly trained how to run a 1-minute reading assessment using provided materials on their child. The other cross-randomized treatment trains school leaders to better coach their teachers in teaching literacy, and also how to interact/engage parents. I am studying the interaction of the exogenous changes in school quality and parental investment at home, and how these work together to impact the reading ability of the students.
Experimental design
I will use a 2 by 2 factorial design where the parent meetings are randomized to half the schools, and school leader training is cross randomized on top of that. So there will be 4 total groups: 1 getting nothing, 1 getting only parent meetings, 1 getting only school leader training, and 1 getting both.
Design details
Randomization method
Done using R in office using the following randomization code and seed.
# Randomize government schools
here::i_am("code/randomize_baseline.R")
pacman::p_load(tidyverse, randomizr)
gov_schools_orf_df = haven::read_dta(here::here(
"data",
"clean_baseline_gov_schools.dta"
))
orf_df = haven::read_dta(here::here("data", "clean_baseline_gov_students.dta"))
set.seed(612873201)
# Create stratification cells: 4 schools per cell
# There are three schools that ran no assessments: they are in a strata of three together
# There are 62 remaining after that, so 2 over 4 per cell. Those are added to cells roughly in the middle
gov_schools_orf_df = gov_schools_orf_df |>
arrange(letter_sound_knowledge_eb) |>
mutate(
stratification_cell = c(
rep(1:6, each = 4), # First 24 schools: 6 strata of 4
rep(7, 4), # Next 4 schools: 1 stratum of 4 (was 5)
rep(8, 4), # Next 4 schools: 1 stratum of 4 (was 5)
rep(9:15, each = 4), # Next 28 schools: 7 strata of 4
rep(16, 3) # Last 3 schools: 1 stratum of 3
)
) |>
mutate(
treat = block_ra(
blocks = stratification_cell
)
) |>
relocate(treat, .after = school) |>
relocate(stratification_cell, .after = treat)
haven::write_dta(
gov_schools_orf_df,
here::here("data", "clean_baseline_gov_schools_with_treatment.dta")
)
# Merge with student level
tmp_schools = gov_schools_orf_df |> select(school, treat, stratification_cell)
orf_df = orf_df |>
left_join(tmp_schools, by = join_by(school)) |>
relocate(treat, .after = student_name) |>
relocate(stratification_cell, .after = treat)
orf_df |>
haven::write_dta(here::here(
"data",
"clean_baseline_gov_students_with_treatment.dta"
))
#### Randomize private schools
private_schools_df = haven::read_dta(here::here(
"data",
"clean_baseline_private_schools.dta"
))
private_students_df = haven::read_dta(here::here(
"data",
"clean_basleine_private_students.dta"
))
set.seed(12398173)
# Firsts randomize 24 private schools that have an observation for beginning of the year assessments
# Create stratification cells: 4 schools per cell
private_schools = private_schools_df |>
filter(!is.na(letter_sound_knowledge_eb)) |>
arrange(letter_sound_knowledge_eb) |>
mutate(
stratification_cell = c(
rep(1:6, each = 4) + 16 # There are 16 strat cells from gov schools
)
) |>
mutate(
treat = block_ra(
blocks = stratification_cell
)
) |>
relocate(treat, .after = school) |>
relocate(stratification_cell, .after = treat)
# Now randomize 16 schools that didn't have an assessment
private_schools_egra = private_schools_df |>
filter(is.na(letter_sound_knowledge_eb)) |>
arrange(letter_sound_knowledge_eb) |>
mutate(
stratification_cell = c(
rep(1:4, each = 4) + 22 # There are 16 strat cells from gov schools and 6 from first batch of private schools
)
) |>
mutate(
treat = block_ra(
blocks = stratification_cell
)
) |>
relocate(treat, .after = school) |>
relocate(stratification_cell, .after = treat)
# Combine
private_schools_combined_df = private_schools |>
bind_rows(private_schools_egra)
haven::write_dta(
private_schools_combined_df,
here::here("data", "clean_baseline_private_schools_with_treatment.dta")
)
# Write treated schools list for Simon
gov_schools_orf_df |>
filter(treat == 1) |>
select(school) |>
writexl::write_xlsx(here::here("data", "treated_gov_schools.xlsx"))
private_schools_combined_df |>
filter(treat == 1) |>
select(school) |>
writexl::write_xlsx(here::here("data", "treated_private_schools.xlsx"))
# Combine into one data set
all_schools_df = gov_schools_orf_df |>
bind_rows(private_schools_combined_df) |>
rename(treat_parent = treat)
### Now cross-randomize school leader training
# This is done using the same stratification cels
# So basically within strat cell 1, the two treat schools are split and the two control schools are split
set.seed(312323814)
cross_randomized_df = map(
seq_along(1:max(all_schools_df$stratification_cell)),
function(i) {
all_schools_df |>
filter(stratification_cell == {{ i }}) |>
mutate(treat_school_leader = block_ra(blocks = treat_parent)) |>
relocate(treat_school_leader, .after = treat_parent)
}
) |>
bind_rows()
# Write to excel
cross_randomized_df |>
haven::write_dta(here::here(
"data",
"full_treatment_list.dta"
))
Randomization unit
Schools.
Clusters
103 schools. 63 government and 40 private.
Observations
Approximately 25 children per school, so 2,575 students.
Arms
51 treatment schools and 52 control.
Primary outcomes
Letter sound knowledge measured by teacher-led oral reading fluency assessments.
Secondary outcomes
For the government schools, I will explore using attendance as an outcome, but not all schools are reliably recording this. Conditional on securing grant funding, I will also measure parental engagement using a survye.
Cited by
No citation data available for this registration yet.