DAAF-Contribution-Community/daaf / .claude/skills/education-data-source-meps
education-data-source-meps
VerifiedMEPS — Urban Institute modeled school-level poverty (% at 100% FPL) for public schools. MEPS 2.0 (released 2025-12-11) covers school years 2009-10 through 2022-23 and is planned for annual updates; Portal year 2022 returned rows while 2023 was empty on 2026-08-06 (re-probed). Use when FRPL is unreliable due to CEP and for consistent cross-state measurement.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 17.5 KB |
| references/comparison-to-frpl.md | markdown | 9.7 KB |
| references/data-quality.md | markdown | 15.6 KB |
| references/data-sources.md | markdown | 11.0 KB |
| references/methodology.md | markdown | 11.4 KB |
| references/variable-definitions.md | markdown | 13.7 KB |
references/data-quality.md
View on GitHub# MEPS Data Quality and Appropriate Uses
Understanding the limitations, uncertainty, and appropriate applications of Model Estimates of Poverty in Schools.
> **Portal Encoding Note:** MEPS data uses **native nulls** for missing values, not the coded values (-1, -2, -3) used by CCD and other sources. Always use null checks rather than negative value filtering.
## Key Data Quality Considerations
### MEPS is a Modeled Estimate
MEPS values are **statistical estimates**, not direct counts:
- Generated from a multilevel (mixed-effects) linear regression model
- Subject to estimation error
- Standard errors quantify uncertainty
- Individual school estimates less reliable than aggregates
**Implication**: Use standard errors for statistical inference; don't treat point estimates as ground truth.
### Estimation Uncertainty (`meps_poverty_se`)
The standard error indicates reliability. Portal values are on a **percentage scale** (matching `meps_poverty_pct`):
| `meps_poverty_se` Range | Interpretation | Typical Context |
|-------------------------|----------------|-----------------|
| < 1.0 | Very reliable | Large schools, abundant data |
| 1.0 - 2.0 | Reliable | Typical schools |
| 2.0 - 3.0 | Moderate uncertainty | Smaller schools, unusual characteristics |
| > 3.0 | High uncertainty | Small schools, missing predictors |
> **Actual observed range:** 0.52 - 3.77 (percentage points). This is on the same scale as `meps_poverty_pct`.
**Using standard errors:**
```python
import polars as pl
# 95% confidence interval
df = df.with_columns([
(pl.col("meps_poverty_pct") - 1.96 * pl.col("meps_poverty_se")).alias("ci_lower"),
(pl.col("meps_poverty_pct") + 1.96 * pl.col("meps_poverty_se")).alias("ci_upper"),
(pl.col("meps_poverty_se") < 2.0).alias("reliable"),
])
```
```r
library(dplyr)
# 95% confidence interval
df <- df |> mutate(
ci_lower = meps_poverty_pct - 1.96 * meps_poverty_se,
ci_upper = meps_poverty_pct + 1.96 * meps_poverty_se,
reliable = meps_poverty_se < 2.0
)
```
## Known Limitations
### 1. Public Schools Only
MEPS covers **only public schools**:
- No private schools
- No religious schools
- No home schools
- Limited alternative school coverage
**Impact**: Cannot compare public vs private school poverty; national totals exclude private sector.
### 2. Time Coverage
| Version | Years Available | Notes |
|---------|-----------------|-------|
| MEPS 1.0 | 2013-14 to 2018-19 (superseded) | Original release |
| MEPS 2.0 | School years 2009-10 through 2022-23 | Released 2025-12-11; Urban states updates will occur annually |
> **Portal status re-probed 2026-08-06:** The Portal's 2022 endpoint returned rows (count 94,941), while its 2023 endpoint returned an empty result (count 0). Portal `year=2022` corresponds to the terminal 2022-23 school year convention used in the release description. Re-probe the needed year before analysis.
**Impact**: Cannot analyze poverty outside the years supplied by the selected release and Portal file.
### 3. Publication Vintage
The terminal year reflects the currently published release. Urban documents an annual update intention for MEPS 2.0 but does not document an inherent or guaranteed fixed lag. The current gap should therefore be described as a **publication-vintage observation**, not as a methodological rule and not as a proven consequence of CCD or SAIPE production timing.
**Impact**: Record the release date and covered school years, and verify that the Portal endpoint or mirror actually contains the target year.
### 4. 100% FPL Only
MEPS measures **only students at or below 100% FPL**:
- Does not capture near-poverty (100-185% FPL)
- Lower threshold than FRPL
- Some economically disadvantaged students not counted
**Impact**: MEPS will show lower poverty rates than FRPL even when FRPL is reliable.
### 5. Model Assumptions
MEPS 2.0 fits a multilevel (mixed-effects) linear regression of adjusted SAIPE district poverty on district-aggregated free-lunch (or direct-certification) shares, with state random intercepts and slopes and district covariates, then applies the parameters to school-level data (Gutierrez & Blagg 2025). It assumes:
- District-level relationships carry over to schools within the district (implied by applying district-level parameters to schools; DAAF reading, not an Urban statement)
- The distribution of students across federal-poverty-level thresholds is relatively consistent across types of schools and districts (Urban's appendix B)
- Available predictors capture relevant variation
**Impact**: Estimates may be biased for schools with unusual characteristics not captured by the model.
### 6. Race and Ethnicity Bias
Urban's bias analysis finds that differences from adjusted SAIPE are larger by district race and ethnicity than by the other district characteristics it examined (which are within about 2 percentage points on average). For every 10 percentage-point increase in a district's Black student share, original MEPS 2.0 underestimates by 0.42 percentage points on average; for every 10-point increase in the Asian student share, it overestimates by 0.31 points (Gutierrez & Blagg 2025):
- Modified MEPS reallocates modeled poor students so each district matches adjusted SAIPE, preserving relative school differences, both up and down
- Urban suggests `meps_mod_poverty_pct` for analyses only of schools within large-enrollment districts, and original `meps_poverty_pct` when smaller districts are included
### 7. Small School Uncertainty
Smaller schools have larger estimation errors:
- Less information for the model
- Higher standard errors
- Consider combining small schools for analysis
## Appropriate Uses
### Strongly Appropriate
| Use Case | Why Appropriate |
|----------|-----------------|
| Cross-state poverty comparison | MEPS designed specifically for this |
| Time trends (2009-10 to 2022-23) | Consistent methodology |
| Poverty-achievement gap analysis | Better control than FRPL |
| Identifying high-poverty schools for research | More comparable than FRPL in CEP era; not for allocating resources within a state or district (Urban) |
| Research on school resources and poverty | Consistent measure |
| District-level aggregation | Only Modified MEPS is adjusted so district aggregates match adjusted SAIPE. Original MEPS is a model prediction |
### Appropriate with Caveats
| Use Case | Caveats |
|----------|---------|
| Individual school analysis | Use standard errors; acknowledge uncertainty |
| Small school analysis | High uncertainty; consider aggregation |
| Analyses limited to large-enrollment districts | Consider `meps_mod_poverty_pct` (Urban); SAIPE margins of error are wide in small districts |
| Year-over-year changes for single school | Changes may be within uncertainty |
| Near-poverty analysis | MEPS only captures 100% FPL |
### Not Appropriate
| Use Case | Why Not | Alternative |
|----------|---------|-------------|
| Private school poverty | Not covered | Use other measures |
| Before 2009-10 | No data | Use FRPL (with caveats) |
| Real-time monitoring | Latest published vintage may not cover the current school year | Use current local data and document the vintage difference |
| 185% FPL threshold | Different threshold | Use FRPL (with CEP adjustment) |
| Meal program planning | Wrong measure | Use FRPL/ISP |
| Compliance with FRPL-based formulas | May not satisfy requirements | Use FRPL |
| Allocating resources within a state or district | Urban: "It should not be used to allocate resources within a state or district" | Use the allocation data required by state/federal rules |
## Statistical Considerations
### Comparing Schools
When comparing poverty between schools:
```python
def statistically_different(pct_a: float, se_a: float, pct_b: float, se_b: float, alpha: float = 0.05) -> bool:
"""Test if two schools have significantly different poverty rates.
Args:
pct_a: meps_poverty_pct for school A
se_a: meps_poverty_se for school A
pct_b: meps_poverty_pct for school B
se_b: meps_poverty_se for school B
alpha: significance level (0.05 or 0.01)
"""
diff = pct_a - pct_b
se_diff = (se_a**2 + se_b**2)**0.5
z_score = abs(diff) / se_diff
z_critical = 1.96 if alpha == 0.05 else 2.58
return z_score > z_critical
```
```r
# Test if two schools have significantly different poverty rates.
# pct_a, se_a: meps_poverty_pct and meps_poverty_se for school A
# pct_b, se_b: meps_poverty_pct and meps_poverty_se for school B
# alpha: significance level (0.05 or 0.01)
diff <- pct_a - pct_b
se_diff <- sqrt(se_a^2 + se_b^2)
z_score <- abs(diff) / se_diff
z_critical <- if (alpha == 0.05) 1.96 else 2.58
statistically_different <- z_score > z_critical
```
### Aggregating to Higher Levels
For district or state aggregation, use enrollment-weighted averages (requires joining with CCD enrollment data first):
```python
import polars as pl
# Enrollment-weighted aggregation with proper SE propagation
district_agg = (
df.filter(
pl.col("meps_poverty_pct").is_not_null()
& pl.col("enrollment").is_not_null()
)
.group_by("leaid")
.agg(
((pl.col("meps_poverty_pct") * pl.col("enrollment")).sum()
/ pl.col("enrollment").sum()).alias("meps_weighted"),
pl.col("enrollment").sum().alias("total_enrollment"),
# SE of weighted average (approximate)
((pl.col("meps_poverty_se").pow(2) * pl.col("enrollment").pow(2)).sum().sqrt()
/ pl.col("enrollment").sum()).alias("meps_se_agg"),
)
)
```
```r
library(dplyr)
# Enrollment-weighted aggregation with proper SE propagation
district_agg <- df |>
filter(!is.na(meps_poverty_pct), !is.na(enrollment)) |>
group_by(leaid) |>
summarise(
meps_weighted = sum(meps_poverty_pct * enrollment, na.rm = TRUE) /
sum(enrollment, na.rm = TRUE),
total_enrollment = sum(enrollment, na.rm = TRUE),
# SE of weighted average (approximate)
meps_se_agg = sqrt(sum(meps_poverty_se^2 * enrollment^2, na.rm = TRUE)) /
sum(enrollment, na.rm = TRUE),
.groups = "drop"
)
```
### Regression with MEPS
When using MEPS as a control variable, note that `meps_poverty_pct` is a modeled estimate with known standard error (`meps_poverty_se`). Consider:
- **Simple approach:** Use `meps_poverty_pct` directly as a regressor (ignores measurement error)
- **Better approach:** Use errors-in-variables regression or sensitivity analysis varying MEPS within its confidence interval
## Data Validation Checks
### Before Using MEPS Data
> **Note:** MEPS uses native nulls, not negative coded values. Adjust validation accordingly.
1. **Check for missing values (MEPS uses nulls, not negative codes)**
```python
import polars as pl
null_pct = df["meps_poverty_pct"].null_count() / len(df)
print(f"Missing: {null_pct:.1%}")
```
```r
library(dplyr)
null_pct <- sum(is.na(df$meps_poverty_pct)) / nrow(df)
cat(paste0("Missing: ", sprintf("%.1f%%", null_pct * 100)), "\n")
```
2. **Verify reasonable ranges**
```python
valid = df.filter(pl.col("meps_poverty_pct").is_not_null())
assert valid["meps_poverty_pct"].min() >= 0, "Negative poverty values"
assert valid["meps_poverty_pct"].max() <= 100, "MEPS out of range"
assert valid["meps_poverty_se"].min() >= 0, "Negative SE values"
```
```r
library(dplyr)
valid <- df |> filter(!is.na(meps_poverty_pct))
stopifnot("Negative poverty values" = min(valid$meps_poverty_pct) >= 0)
stopifnot("MEPS out of range" = max(valid$meps_poverty_pct) <= 100)
stopifnot("Negative SE values" = min(valid$meps_poverty_se, na.rm = TRUE) >= 0)
```
3. **Check coverage**
```python
print(f"Schools covered: {df['ncessch'].n_unique():,}")
print(f"States covered: {df['fips'].n_unique()}")
print(f"Years covered: {df['year'].unique().sort().to_list()}")
```
```r
library(dplyr)
cat(sprintf("Schools covered: %s\n", format(n_distinct(df$ncessch), big.mark = ",")))
cat(sprintf("States covered: %d\n", n_distinct(df$fips)))
cat(sprintf("Years covered: %s\n", paste(sort(unique(df$year)), collapse = ", ")))
```
4. **Assess reliability distribution**
```python
print(df["meps_poverty_se"].describe())
reliable_pct = df.filter(pl.col("meps_poverty_se") < 2.0).height / len(df)
print(f"Reliable estimates (SE<2.0): {reliable_pct:.1%}")
```
```r
library(dplyr)
print(summary(df$meps_poverty_se))
reliable_pct <- sum(df$meps_poverty_se < 2.0, na.rm = TRUE) / nrow(df)
cat(sprintf("Reliable estimates (SE<2.0): %.1f%%\n", reliable_pct * 100))
```
## Reporting Recommendations
### In Research Papers
Always report:
1. MEPS version used
2. Years included
3. Sample restrictions applied
4. How standard errors were used
5. Acknowledgment that MEPS measures 100% FPL (not 185%)
Example text:
> "We measure school-level poverty using the Urban Institute's Model Estimates of Poverty in Schools (MEPS), which estimates the share of students from households with incomes at or below 100 percent of the federal poverty level. MEPS provides a consistent poverty measure across states and time, unlike FRPL data which is affected by policy variation and Community Eligibility Provision adoption (Gutierrez, Blagg, & Chingos, 2022; Gutierrez & Blagg, 2025)."
Full references for the example:
- Gutierrez, Emily, Kristin Blagg, and Matthew Chingos. 2022. *Model Estimates of Poverty in Schools: A New School-Level Measure of Economic Need.* Washington, DC: Urban Institute.
- Gutierrez, Emily, and Kristin Blagg. 2025. *Model Estimates of Poverty in Schools 2.0.* Washington, DC: Urban Institute.
### In Policy Reports
Clarify:
1. Difference between MEPS and FRPL
2. Why MEPS is more comparable
3. Limitation that it measures 100% FPL
4. Data availability and recency
## Common Pitfalls
### 1. Ignoring Standard Errors
**Problem**: Treating MEPS point estimates as exact values
**Solution**: Always consider `meps_poverty_se` in comparisons and conclusions
### 2. Conflating with FRPL
**Problem**: Assuming MEPS and FRPL are interchangeable
**Solution**: Clearly distinguish; note different thresholds and methodologies
### 3. Single-School Conclusions
**Problem**: Making strong claims about individual schools
**Solution**: Use aggregations or acknowledge uncertainty
### 4. Publication Vintage
**Problem**: Using MEPS for current policy without recording the release date and terminal school year
**Solution**: State the release and years covered; verify the target Portal year and describe any gap as publication vintage
### 5. Ignoring Modified MEPS
**Problem**: Choosing between original and Modified MEPS without considering district size
**Solution**: Urban suggests `meps_mod_poverty_pct` for analyses only of schools within large-enrollment districts, and original `meps_poverty_pct` when smaller districts are included
## Quality Assurance Checklist
Before publishing analysis using MEPS:
- [ ] Documented MEPS version and years used
- [ ] Excluded or flagged missing/suppressed values
- [ ] Considered standard errors in key findings
- [ ] Noted 100% FPL threshold (different from FRPL)
- [ ] Acknowledged limitations in discussion
- [ ] Used appropriate aggregation methods
- [ ] Cited Urban Institute methodology report
## Summary
MEPS is a **high-quality, research-grade** poverty measure when used appropriately:
| Strength | Limitation |
|----------|------------|
| Cross-state comparable | Model-based (not direct counts) |
| Time-consistent | Published release has a finite terminal school year; annual updates are stated but not guaranteed on a fixed lag |
| Fit to adjusted Census SAIPE | 100% FPL only (not 185%) |
| Accounts for CEP | Public schools only |
| Standard errors provided | Some uncertainty for small schools |
**Bottom line**: MEPS is the best available school-level poverty measure for research requiring cross-state or temporal comparability. Use standard errors, acknowledge limitations, and choose appropriate use cases.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |