Spend-versus-outcome efficiency frontier (free-disposal hull)
Source:R/tges_analysis.R
tges_frontier.RdWhere tges_efficiency() only labels a quadrant, this
computes a proper efficiency score against the free-disposal-hull frontier.
For each district it finds the peer that achieves at least the same outcome for
the least spend, and reports the ratio (0-1, 1 = on the frontier) plus that
reference district and the dollars overspent relative to it. This answers the
MarGrady brief's open cost-effectiveness question with a number, not a
quadrant: "Bayonne hits your graduation rate for $2,100/pupil less; you are at
0.81 efficiency."
Usage
tges_frontier(
spend_df,
outcome_df,
spend_col = "Per Pupil costs",
outcome_col,
peer = c("tges_group", "dfg", "county", "statewide"),
year_col = "end_year",
dfg_revision = 2000
)Arguments
- spend_df
A tidied TGES table (one report year, one
calc_type) carryingdistrict_idandspend_col.- outcome_df
A district-level frame with
district_id, the year column, andoutcome_col(higher = better).- spend_col
Character. Spend column in
spend_df. Default"Per Pupil costs".- outcome_col
Character. Outcome column in
outcome_df.- peer
Character. Peer group for the frontier. One of
"tges_group"(default),"dfg","county","statewide".- year_col
Character. Year column, present in both frames. Default
"end_year".- dfg_revision
Numeric. DFG revision when
peer = "dfg". Default 2000.
Value
A tibble: entity columns, peer_group, the year, spend,
outcome, efficiency_score, on_frontier,
reference_district_id, reference_district_name,
reference_spend, and excess_spend.
Details
The free-disposal hull (FDH) is the dependency-free efficiency
frontier: a district is efficient if no peer in the same group does at least as
well on the outcome for strictly less spend. The input-oriented score is
min(spend among peers with outcome >= yours) / your spend. It needs no
linear-programming solver and makes no functional-form assumption (unlike the
regression residual in tges_efficiency()). Spend is treated as the input
(lower is better) and the outcome as the output (higher is better); pass an
outcome where higher is better (a rate or a percentile).
Cautions. Use one consistent per-pupil definition for spend_col
(CSG1 budgetary vs CSG1AA total), filter spend_df to one
calc_type, and make sure the outcome and spend peer systems are stated.
The frontier is only as meaningful as the peer group: with few districts per
group, most land on the frontier by default.
Examples
if (FALSE) { # \dontrun{
library(dplyr)
grate <- fetch_grad_rate(2023, methodology = "4 year") %>%
add_dfg() %>%
filter(dfg == "A", is_district, subgroup == "total") %>%
grate_percentile_rank(peer_type = "dfg")
spend <- fetch_tges(2024)$CSG1 %>%
filter(calc_type == "Actuals", end_year == 2023)
fr <- tges_frontier(spend, grate, outcome_col = "grad_rate_percentile",
peer = "dfg")
fr %>% filter(district_id == "3570") %>%
select(district_name, efficiency_score, reference_district_name, excess_spend)
} # }