Synthetic NSECE 2019 Center-Based Provider Data

A synthetic dataset mimicking the structure and statistical properties of the 2019 National Survey of Early Care and Education (NSECE) center-based provider data. Generated via Gaussian copula to preserve marginal distributions and rank-correlation structure of the original restricted-use data.

Usage

nsece_synth

Format

A data frame with approximately 6,785 rows and 13 columns:

provider_id: Integer. Unique provider identifier (1 to N).
state_id: Integer. State identifier (1 to 51, including DC).
y: Integer. Infant/toddler (IT) enrollment count. Zero for centers not serving IT (z = 0).
n_trial: Integer. Total enrollment for children ages 0–5. Always \(\ge 1\).
z: Integer. Participation indicator: 1 if the center serves any IT children, 0 otherwise. Equal to I(y > 0).
it_share: Numeric. IT enrollment share y / n_trial. Zero when z = 0.
poverty: Numeric. Community poverty rate (percent below poverty line). Range approximately 2–54.
urban: Numeric. Community urbanization rate (percent urban population). Range 0–100, heavily right-skewed.
black: Numeric. Community percent Black population. Range 0–97.
hispanic: Numeric. Community percent Hispanic population. Range 0–98.
weight: Numeric. Survey sampling weight. Positive values; larger weights indicate the provider represents more units in the population.
stratum: Integer. Sampling stratum identifier.
psu: Integer. Primary sampling unit (PSU) identifier, nested within strata.

Source

Generated from the NSECE 2019 restricted-use data via Gaussian copula (Miratrix, 2025) with hurdle Beta-Binomial outcomes calibrated to the empirical moments of the original data. See data-raw/generate_synthetic.R for the full generation script.

Details

The data-generating process uses a two-part hurdle Beta-Binomial model:

Part 1 (Extensive margin): Whether a center serves IT children at all, modeled as \(z_i \sim \textrm{Bernoulli}(q_i)\) where \(\textrm{logit}(q_i) = X_i \alpha + \delta_{1,s_i}\).
Part 2 (Intensive margin): Among servers, the IT enrollment count follows a zero-truncated Beta-Binomial: \(y_i \mid z_i = 1 \sim \textrm{ZT-BetaBin}(n_i, \mu_i, \kappa)\) where \(\textrm{logit}(\mu_i) = X_i \beta + \delta_{2,s_i}\).

The structural zero rate is approximately 35 percent, and the mean IT share among servers is approximately 48 percent. These match the empirical moments of the original NSECE 2019 data.

References

National Survey of Early Care and Education (NSECE), 2019. U. S. Department of Health and Human Services.

Examples

data(nsece_synth)
head(nsece_synth)
#>   provider_id state_id  y n_trial z   it_share  poverty    urban      black
#> 1           1       33  1      24 1 0.04166667 29.70972 99.95720  4.7455889
#> 2           2       37 30      58 1 0.51724138 19.30702 99.96461  4.5828925
#> 3           3       26  0     101 0 0.00000000 12.71033 99.92934 18.5843027
#> 4           4        5  0      19 0 0.00000000 18.54633 15.12154  0.8891587
#> 5           5        5 31     125 1 0.24800000 10.63834 99.94319  1.5584889
#> 6           6       10  0     172 0 0.00000000 19.32121 99.96716 32.6650238
#>    hispanic    weight stratum psu
#> 1 60.887394  3.276547       3 138
#> 2 42.464795  6.939392      27 402
#> 3  6.466195  4.953660      20 367
#> 4  9.839974 16.319319       1  42
#> 5  9.304892  5.363512       1  45
#> 6 20.409599  2.164282       5 202

# Basic summary
cat("N =", nrow(nsece_synth), "\n")
#> N = 6785 
cat("Zero rate:", round(1 - mean(nsece_synth$z), 3), "\n")
#> Zero rate: 0.354 
cat("IT share (servers):",
    round(mean(nsece_synth$it_share[nsece_synth$z == 1]), 3), "\n")
#> IT share (servers): 0.496 

# State distribution
cat("States:", length(unique(nsece_synth$state_id)), "\n")
#> States: 51