This source did not publish a separate summary. Review SKILL.md before using the skill.
SKILL.md
Scorecard Data Source Reference
College Scorecard — the primary institutional-level source for post-enrollment labor market outcomes, linking NSLDS financial aid records to IRS/Treasury earnings data. Use when comparing institutions on post-graduation earnings, loan repayment, or student debt, or when actual tax-record-based earnings are required rather than survey estimates. Covers six sub-datasets accessed via Portal mirrors. Critical limitation: tracks only Title IV federal aid recipients, not all students.
Federal data on post-college outcomes including earnings, debt, and repayment for students who received Title IV financial aid. Links education records to IRS tax data for actual earnings, making it the primary source for post-college labor market outcomes.
CRITICAL: Value Encoding and Missing Data
The Education Data Portal uses integer encodings for all categorical variables
and lowercase, restructured variable names that differ from the original
Scorecard column names. Suppression encoding differs by dataset:
Earnings/counts: -3 integer code is the primary suppression indicator
Yes/No flags (institutional characteristics): null for missing, 0/1 for valid
Rates (repayment, default): null for missing
The original Scorecard string "PrivacySuppressed" does NOT appear in Portal data
Context
pred_degree_awarded_ipeds
HBCU / tribal flags
religious_affiliation
Portal (integer)
0-4
0 / 1
Integer codes 22-200
Original Scorecard
String labels
String labels
String labels
See ./references/variable-definitions.md for complete encoding tables.
What is College Scorecard?
Publisher: U.S. Department of Education
Primary value: Post-college labor market outcomes (earnings) and debt/repayment metrics
Data sources: NSLDS (loans/aid), IRS/Treasury (earnings), IPEDS (institutional characteristics)
Coverage: Title IV federal aid recipients only — not all students
Unique feature: Links education to IRS tax records for actual earnings data
Access: Education Data Portal mirrors (parquet/CSV); see datasets-reference.md for paths, mirrors.yaml for mirror config, fetch-patterns.md for fetch code
Primary identifier: unitid (IPEDS institution ID)
Reference File Structure
File
Purpose
When to Read
earnings-data.md
Post-college earnings methodology, cohorts, time horizons
Analyzing earnings outcomes
debt-repayment.md
Student debt, repayment rates, default rates
Analyzing debt or loan outcomes
completion-rates.md
Completion metrics vs IPEDS
Comparing graduation rates
population-coverage.md
Title IV limitation details, who is included/excluded
Understanding data representativeness
variable-definitions.md
Key variables, naming conventions, special values
Building queries or interpreting results
data-quality.md
Suppression rules, selection bias, known limitations
Assessing data reliability
field-of-study.md
Program-level earnings and debt data
Analyzing outcomes by major/CIP code
Decision Trees
What outcome am I researching?
Outcome type?
├─ Post-college earnings
│ ├─ Institution-level → ./references/earnings-data.md
│ └─ By field of study → ./references/field-of-study.md
├─ Student debt levels
│ ├─ Cumulative borrowing → ./references/debt-repayment.md
│ └─ Debt by field → ./references/field-of-study.md
├─ Loan repayment/default
│ └─ Repayment rates → ./references/debt-repayment.md
├─ Completion rates
│ └─ Scorecard completion → ./references/completion-rates.md
└─ Understanding limitations
├─ Who is included → ./references/population-coverage.md
└─ Data quality issues → ./references/data-quality.md
How do I interpret this data?
Interpretation question?
├─ Why are earnings suppressed?
│ └─ Privacy thresholds → ./references/data-quality.md
├─ What does "6-year earnings" mean?
│ └─ Cohort timing → ./references/earnings-data.md
├─ Why don't Scorecard rates match IPEDS?
│ └─ Different cohorts → ./references/completion-rates.md
├─ What loans are included in debt?
│ └─ Federal only → ./references/debt-repayment.md
└─ How representative is this data?
└─ Title IV coverage → ./references/population-coverage.md
Building a query?
Query construction?
├─ Variable names and codes → ./references/variable-definitions.md
├─ Suppression flags to handle → ./references/data-quality.md
├─ Understanding cohort years → ./references/earnings-data.md
└─ Field-level queries → ./references/field-of-study.md
Quick Reference: Scorecard Variables
Portal Data Structure (CRITICAL)
The Portal uses LONG format with time horizon as a column, NOT the WIDE format from original Scorecard bulk download files. Portal column names are all lowercase and differ significantly from original Scorecard names.
Original Scorecard (WIDE)
Portal Column (LONG)
How to Get
MD_EARN_WNE_P6
earnings_med
Filter: years_after_entry == 6
MD_EARN_WNE_P10
earnings_med
Filter: years_after_entry == 10
COUNT_WNE_P6
count_working
Filter: years_after_entry == 6
MN_EARN_WNE_P6
earnings_mean
Filter: years_after_entry == 6
CONTROL, INSTNM
NOT IN EARNINGS
Join to IPEDS directory or inst_characteristics dataset
CDR3
default_rate
In repayment_fsa dataset; filter: years_since_entering_repay
RPY_3YR_RT
repay_rate
In repayment_nslds dataset; filter: years_since_entering_repay == 3
Earnings Dataset Columns (Actual Portal Names)
Source dataset:scorecard/colleges_scorecard_earnings (203,066 rows x 33 columns)
Portal Column
Type
Description
Original Scorecard
unitid
Int64
IPEDS institution ID
UNITID
opeid
String
OPE ID (8-digit, zero-padded)
OPEID
year
Int64
Data year (2003-2014, 2018)
File year
years_after_entry
Int64
Years since first enrollment (6-10)
Encoded in variable name
cohort_year
Int64
Entry cohort year
Encoded in variable name
earnings_med
Int64
Median earnings (W-2)
MD_EARN_WNE_P*
earnings_mean
Int64
Mean earnings
MN_EARN_WNE_P*
earnings_sd
Int64
Standard deviation of earnings
SD_EARN_WNE_P*
earnings_pct10
Int64
10th percentile earnings
PCT10_EARN_WNE_P*
earnings_pct25
Int64
25th percentile earnings
PCT25_EARN_WNE_P*
earnings_pct75
Int64
75th percentile earnings
PCT75_EARN_WNE_P*
earnings_pct90
Int64
90th percentile earnings
PCT90_EARN_WNE_P*
count_working
Int64
Count working and not enrolled
COUNT_WNE_P*
count_not_working
Int64
Count not working and not enrolled
COUNT_NWNE_P*
earnings_greater_than_25k_pct
Float64
Share earning > $25K
GT_25K_P*
earnings_lowinc_mean
Int64
Mean earnings, low-income
MN_EARN_WNE_INC1_P*
earnings_midinc_mean
Int64
Mean earnings, mid-income
MN_EARN_WNE_INC2_P*
earnings_highinc_mean
Int64
Mean earnings, high-income
MN_EARN_WNE_INC3_P*
earnings_dep_mean
Int64
Mean earnings, dependent students
—
earnings_dep_lowinc_mean
Int64
Mean earnings, dependent low-income
—
earnings_ind_mean
Int64
Mean earnings, independent students
—
earnings_female_mean
Int64
Mean earnings, female
—
earnings_male_mean
Int64
Mean earnings, male
—
count_working_*
Int64
Count working by subgroup
—
Key Identifiers
ID
Format
Level
Example
Notes
unitid
6-digit integer
Institution
110635
Same as IPEDS unitid; primary join key
opeid
8-digit string
OPE (Title IV)
"00100200"
Zero-padded; present in all datasets
opeid6
Integer
6-digit OPE
1002
Numeric, no zero-padding
Data Timing
Metric
Dimension Column
Values
Typical Lag
Earnings
years_after_entry
6, 7, 8, 9, 10
Data from 7+ years ago
Default
years_since_entering_repay
2, 3
Varies
Repayment
years_since_entering_repay
1, 3, 5, 7
Varies
"After entry" means after first enrollment, not after graduation.
Categorical Value Encodings (Institutional Characteristics Dataset)
Datasets for Scorecard are available via the mirror system. See datasets-reference.md for canonical paths, mirrors.yaml for mirror configuration, and fetch-patterns.md for fetch code patterns.
Codebooks are .xls files co-located with data in all mirrors. Use get_codebook_url() from fetch-patterns.md to construct download URLs.
Truth Hierarchy: When interpreting variable values, apply this priority:
Actual data file (what you observe in the parquet/CSV) — this IS the truth
Live codebook (.xls in mirror) — authoritative documentation, may lag
This skill documentation — convenient summary, may drift from codebook
If this documentation contradicts the codebook, trust the codebook. If the codebook contradicts observed data, trust the data and investigate.
Scorecard naming note: Data file paths differ significantly from codebook paths. Notable mismatches: data repayment_fsa vs codebook default; data inst_characteristics vs codebook institutional-characteristics; data repayment_nslds vs codebook repayment; data student_body_nslds vs codebook student-characteristics_aid-applicants; data student_body_treasury vs codebook student-characteristics_home-neighborhood. Always use the exact paths shown above.
Fetching Data
import polars as pl
from fetch_utils import fetch_from_mirrors # See fetch-patterns.md
# Fetch earnings data
earnings = fetch_from_mirrors("scorecard/colleges_scorecard_earnings")
# Filter by time horizon (LONG format — filter, don't use wide column names)
six_yr = earnings.filter(pl.col("years_after_entry") == 6)
# Filter for valid earnings (exclude -3 suppression code)
valid = six_yr.filter(
(pl.col("earnings_med").is_not_null()) &
(pl.col("earnings_med") != -3)
)
# Institution names/control are NOT in the earnings dataset.
# Join to inst_characteristics or IPEDS directory:
inst = fetch_from_mirrors("scorecard/colleges_scorecard_inst_characteristics",
years=[2020])
valid = valid.join(
inst.select("unitid", "inst_name", "pred_degree_awarded_ipeds"),
on="unitid", how="left"
)
Common Pitfalls
Pitfall
Issue
Solution
"All graduates" claims
Scorecard covers Title IV recipients only, not all students
Note Title IV limitation prominently in any analysis
Wage comparison
Comparing to BLS wages or Census income uses different populations
Use for relative comparisons, not absolute claims; document population differences
Ignoring suppression
Many programs have no data due to privacy thresholds
Check suppression rates before analyzing; document coverage
Time lag ignored
Earnings reflect old cohorts (6-year = data from 7+ years ago)
Document data vintage and cohort years explicitly
Total borrowing assumption
Scorecard debt includes only federal loans, not private
State "federal loans only" when reporting debt figures
String codes from docs
Original Scorecard uses string labels; Portal uses integers
Verify actual data types in Portal parquet files; use integer codes
Wide-format variable names
Using MD_EARN_WNE_P10 column name on Portal data
Portal uses LONG format — filter years_after_entry instead
Assuming null = suppressed
Earnings dataset uses -3 for suppression, not null
Filter both: is_not_null() & != -3
Using uppercase names
Original Scorecard uses MD_EARN_WNE_P6; Portal uses earnings_med
Always use lowercase Portal names from actual data
Critical Limitation: Title IV Recipients Only
The single most important caveat for all Scorecard analysis:
Scorecard tracks ONLY students who received federal financial aid (Title IV):
Pell Grants
Federal student loans (Direct, Perkins, PLUS)
Federal work-study
Excluded Group
Impact
Full-pay students
Often higher-income; different outcomes
Students with only state/institutional aid
Missing from data
International students
Not eligible for federal aid
Some graduate students
If they received no federal aid
Coverage varies dramatically by institution type:
Institution Type
Typical Title IV Coverage
For-profit colleges
80-90%+
Community colleges
60-80%
Public flagships
50-70%
Selective private colleges
30-50%
Data systematically overrepresents lower-income students who are more likely to need federal aid.
