Skip to content
Analyst work

Data study · October 2026

Campus rec and GPA

Question

Do students who use the rec center keep better grades and stay enrolled? Chico State is the example.

Headline finding

At the campuses that measured it, yes. At West Virginia University, rec users came back for a second year at a rate 7.1 points above similar students who didn’t use it. Accounting for prior grades and family background didn’t shrink the gap in any of the 4 tests. Whether the rec center causes any of it is still open. The one randomized trial found a small grade gain in its first cohort and none in its second, and nothing in the data measures Chico State students.

My recommendation

Invite at-risk first-years to the WREC by lottery, so Chico State learns whether rec use keeps them enrolled before it spends on more.

Tools
  • JavaScript on Node for the derived rates
  • D3.js for the first dashboard
  • React and SVG for the charts on this page

The question

The client

The recreation and student affairs office at Chico State, home of the Wildcat Recreation Center (WREC). IPEDS puts full-time first-year retention at 82% for the fall 2023 cohort. Set against the 2,166 first-time freshmen in CSU’s 2023-24 census, that is about 390 students who do not come back for a second year.

The decision they face

Should the office spend on the rec center as a way to keep first-years enrolled, and if so, on which students and with what proof that it works?

The data, 2 tables from 28 sources

  • metrics.csv102 rows

    102 figures on rec use, GPA and retention, one per row. They cover campus studies at Purdue, UCF, Michigan State, NC State and West Virginia, a randomized trial at the University of St. Gallen, national student surveys and 10 rows on Chico State.

    Where it came from
    Journal articles, preprints, a dissertation, university news releases, ACHA-NCHA executive summaries, IPEDS College Navigator, CSU’s admissions profile and Chico State’s own pages.
    Why I chose it
    Each figure gets an ID, a source URL and a label for who produced it, so the code and the second table can point at the exact row.
    What it cannot tell us
    Anything about how Chico State students use the WREC. The 10 Chico rows are campus totals. 60 of the 102 figures come from campus rec departments. The WVU and NC State coefficients are the papers’ printed results, so the student records and the code that produced them are not in the repo.
  • variables.csv9 rows

    9 variables that could explain the gap or change its size, each scored 0 to 2 for bias risk.

    Where it came from
    My own coding of metrics.csv, with the IDs of the rows each score rests on.
    Why I chose it
    It puts the confounders in one place and says which ones a study controlled for.
    What it cannot tell us
    Stand as measurement. The scores are judgment, and the evidence IDs are there so a reader can disagree. Its source column points back to metrics.csv, so these rows have no URL of their own.

The 28 sources behind metrics.csv

metrics.csv holds figures from 28 sources on 21 pages. Some pages appear under more than one name, such as a paper listed once per table. Each one is listed with the rows it supplies, the study they belong to and who produced them. 60 of the 102 rows come from campus rec departments.

Show the 28 sources

The process

  1. One row per figure

    Each figure went into a row with its study, a source name, a URL, the date I retrieved it, who produced it and a flag for student self-report. That came to 102 sourced rows, plus 9 rows of my own coding.

  2. One scale

    The 4 tests in finding 1 come in three units. GPA slopes and percentage-point gaps divide as they are. Odds ratios needed a log first, and then I divided each controlled estimate by its raw one.

  3. Join

    variables.csv points at metrics.csv by row ID. The 9 variables cite 27 rows between them, and every ID resolves to a row.

  4. Assumptions

    The trial’s second cohort is reported as no improvement, and the table enters it as 0. The table holds WVU’s raw retention rates as whole percents, so the 7-point raw gap could be anywhere from 6 to 8. That puts the matched share between 89% and 118%.

  5. Check

    The 8 WVU visit bands rise at every step, for retention and for GPA. The four ACHA stress shares add to 100.0%.

Six source rows before and after they become three tests.

Before, six rows of metrics.csv (five of twelve columns shown)
metric_id,study,metric,value,unit
S19,NC State Sanderson 2017,Odds ratio of persisting or graduating per weekly contact hour: no controls,1.49,odds ratio
S21,NC State Sanderson 2017,Odds ratio of persisting or graduating per weekly contact hour: plus SAT and high school GPA,1.53,odds ratio
S24,WVU Zegre 2020,First-year retention: rec users (raw),83,percent
S25,WVU Zegre 2020,First-year retention: non-users (raw),76,percent
S26,WVU Zegre 2020,Retention gap with full controls (logit marginal effect),8.4,percentage points
S27,WVU Zegre 2020,Retention gap in matched sample (coarsened exact matching),7.1,percentage points
After, one row per test
[
  {
    "test": "NC State persistence odds",
    "raw": 0.399,
    "controlled": 0.425,
    "pct_left": 106.6
  },
  {
    "test": "WVU retention, regression",
    "raw": 7,
    "controlled": 8.4,
    "pct_left": 120
  },
  {
    "test": "WVU retention, matched",
    "raw": 7,
    "controlled": 7.1,
    "pct_left": 101.4
  }
]

The NC State pair is an odds ratio, so both sides are logged before dividing. The WVU raw gap is one retention rate minus the other.

The code for the key steps

Each controlled estimate as a share of its raw one, with odds ratios on the log scale. derive.mjs
export function selection(metrics) {
  const v = pick(metrics);
  const rows = [
    { id: "nc_gpa", test: "NC State GPA slope", raw: v("S16"), controlled: v("S18") },
    { id: "nc_odds", test: "NC State persistence odds", raw: Math.log(v("S19")), controlled: Math.log(v("S21")) },
    { id: "wvu_logit", test: "WVU retention, regression", raw: v("S24") - v("S25"), controlled: v("S26") },
    { id: "wvu_matched", test: "WVU retention, matched", raw: v("S24") - v("S25"), controlled: v("S27") },
  ];
  return rows.map((r) => ({ ...r, pct_left: (r.controlled / r.raw) * 100 }));
}
How far rounding in WVU’s raw rates could move that share. derive.mjs
export function roundingBand(rows) {
  return rows
    .filter((r) => r.id.startsWith("wvu"))
    .map((r) => ({ id: r.id, test: r.test, raw: r.raw, raw_lo: r.raw - 1, raw_hi: r.raw + 1, pct_left_lo: (r.controlled / (r.raw + 1)) * 100, pct_left_hi: (r.controlled / (r.raw - 1)) * 100 }));
}

Dead ends

  • The trial against a campus slope

    I put NC State’s slope and the St. Gallen trial on one scale, grade gain in SD per extra weekly session. NC State comes to 0.09. The trial’s 95% interval runs from -0.09 to 1.17, so the trial can neither confirm the slope nor rule it out. The comparison also treats a rec visit and a contact hour as one unit.

  • Chico State’s own retention figure

    Chico State’s institutional research page shows a 74% retention rate and names no cohort. I could not tell which students it covers, so the study uses the IPEDS figure of 82%.

  • A count of students kept

    The first dashboard turned the WVU gap into a number of Chico State students kept each year. That needs a 20% share of Pell first-years who become regular users and a 50% discount for motivation, and no row supplies either. It also reads the WVU gap as an effect of going. I left it off this page.

The 3 findings that change the decision

  1. Finding 1

    104%

    Share of the gap left after accounting for background, median of 4 tests

    In 4 tests from 2 campuses, accounting for prior grades and family background didn’t shrink the gap. What was left came to 100% to 120% of the original gap, with a median of 104%.

    Dataset metrics.csv
    Query derive.mjs, selection()

    Accounting for background left a median 104% of the gap in place

    Controlled estimate as a percent of the raw estimate, so 100% means the controls changed nothing. Both campuses controlled for traits they had on record, and neither measured motivation. WVU’s raw rates are whole percents, so its matched share could sit between 89% and 118%.

    Controlled estimate as a share of raw (%)

    Source: metrics.csv, derive.mjs, selection().

    Show the numbers as a table
    TestControlled estimate as a share of raw (%)Raw estimateControlled estimateRows
    NC State GPA slope100%0.06 GPA points per weekly hour0.06 GPA points per weekly hourS16, S18
    NC State persistence odds107%Odds ratio 1.49, log 0.40Odds ratio 1.53, log 0.43S19, S21
    WVU retention, regression120%7 points, 83% against 76%8.4 pointsS24, S25, S26
    WVU retention, matched101%7 points, 83% against 76%7.1 pointsS24, S25, S27
  2. Finding 2

    0.14 SD

    Grade gain from the incentive in standard deviations, first cohort of 567 students

    In the one randomized trial in the table, a cash incentive raised on-campus sports visits 47% and grades 0.14 standard deviations in the first cohort, a small gain. The second cohort showed no grade gain.

    Dataset metrics.csv
    Query derive.mjs, ittPooled()

    The one trial found a small grade gain in its first cohort and none in its second

    University of St. Gallen, Switzerland. Students were randomized to a cash incentive for on-campus sports, and the first two bars compare everyone offered the incentive with everyone who was not. The second cohort’s value is the table’s coding of “no improvement”.

    Grade effect of the incentive (standard deviations)

    Source: metrics.csv, derive.mjs, ittPooled().

    Show the numbers as a table
    CohortGrade effect of the incentive (standard deviations)Students randomizedWhere the value comes from
    First cohort, 20130.14567Row R03, as reported
    Second cohort, 20140.00754Row R04, reported as no improvement and entered as 0
    Both cohorts, weighted by students0.061,321Weighted average of rows R03 and R04
  3. Finding 3

    12.4 points

    Retention gap over similar non-users, WVU students with a high school GPA under 3.5

    At WVU, rec users who entered with a high school GPA under 3.5 came back at a rate 12.4 points above similar students who didn’t use it. For students at 3.5 or higher the gap was 2.2 points.

    Dataset metrics.csv
    Query derive.mjs, subgroups()

    The WVU gap is 12.4 points for lower-GPA entrants and 2.2 for the rest

    First-year retention of rec users minus matched non-users at West Virginia University, entering classes 2014-15 to 2016-17. Matching balances the traits WVU had on record. Motivation was never on record, so a gap can overstate what going does.

    Matched retention gap (percentage points)

    Source: metrics.csv, derive.mjs, subgroups().

    Show the numbers as a table
    Student groupMatched retention gap (percentage points)GroupRow
    High school GPA under 3.512.4At-risk groupS39
    High school GPA 3.5 or higher2.2Comparison groupS40
    First-generation10.8At-risk groupS38
    Not first-generation6.2Comparison groupS60
    Pell grant10.5At-risk groupS37
    No Pell grant5.7Comparison groupS59

My recommendation

Target at-risk first-years, and pick who goes first by lottery.

  1. A

    Target at-risk first-years

    Advisors and EOP staff invite Pell and first-generation first-years, along with students who entered with a high school GPA under 3.5. A drop in WREC entries triggers a check-in.

    Evidence
    Finding 3. Gaps over similar non-users run 10.5 points for Pell students and 12.4 for lower-GPA entrants. At Chico State, 41% of full-time beginning students hold a Pell grant.
    Main risk
    These gaps come from one campus, and matching on records can’t account for motivation. An invited student may gain less than the gap suggests.
  2. B

    Randomize the rollout and measure

    Hold a lottery for which eligible students get the first invitation. Link WREC entries to institutional research records and compare return rates by group.

    Evidence
    Finding 2. The one trial in the table raised grades 0.14 standard deviations in its first cohort and not at all in its second, and it ran at a Swiss university. The Chico State catalog describes a hand reader at the WREC door, so entries can be tied to students.
    Main risk
    A lottery can come back unclear. The St. Gallen trial randomized 567 students in its first cohort, and its 95% interval for the grade effect per weekly visit still ran from -0.09 to 1.17 standard deviations.
  • Start where the gaps are largest. Students who entered with a high school GPA under 3.5 showed the biggest matched retention gap in finding 3, and first-generation and Pell students came next.
  • Those gaps come from matched comparisons at one campus, and the one randomized trial lost its grade effect in its second cohort. A lottery for the first invitations gives Chico State an answer of its own.
  • The hand reader at the WREC door already logs entries, so return rates can be compared by group with no new equipment.

Limits and next steps

What the data cannot support

  • Cause. Every gap except the trial’s compares students who chose the rec center with students who did not. I scored motivation at 2, the highest bias risk, because no study here measures it. That score is my judgment.
  • Chico State. The 10 Chico rows are campus totals. None links a WREC entry to a grade or to a student coming back, so every gap on this page comes from another campus.
  • Independence. 60 of the 102 figures come from campus rec departments studying their own programs, and 2 campuses supply all 4 tests in finding 1.
  • Retention from the trial. It paid students at one Swiss university to use on-campus sports, and the rows I hold from it are grades.
  • One GPA gap. The raw gaps across campuses run from 0.10 to 0.27 GPA points, and each campus defines a rec user its own way.

What I would test with more access

  • With WREC entry records joined to enrollment records, I would rerun the WVU comparison on Chico State first-years, raw and matched, for retention and for first-year GPA.
  • I would add where a student lives as a control. At WVU 5% of rec users lived off campus against 22% of non-users, and my table marks housing as only partly controlled.
  • I would ask Institutional Research which cohort the 74% on its page describes.

TL;DR

  1. Studied

    What rec center use does for grades and staying enrolled, read from other campuses for Chico State.

  2. Found

    At WVU, rec users came back for a second year at a rate 7.1 points above similar students who didn’t use it.

  3. By the numbers

    WVU retention gap over similar non-users
    7.1 points
    Of the gap left after accounting for background
    104%
    Grade gain in standard deviations, the one trial’s first cohort
    0.14 SD
  4. The catch

    Every gap except the trial’s compares students who chose the rec center with students who did not.

  5. Next

    Join WREC entry records to enrollment and rerun the WVU comparison on Chico State first-years.