{
  "id": "2006-july-august-finding-diamonds-in-the-rough",
  "slug": "2006-july-august-finding-diamonds-in-the-rough",
  "url": "https://dwillis.github.io/uplink/article/2006-july-august-finding-diamonds-in-the-rough.html",
  "year": 2006,
  "month": "July-August",
  "monthNum": 7,
  "headline": "Finding diamonds in the rough",
  "kicker": "SCHOOL DATA",
  "author_name": "MaryJo Sylwester",
  "author_title": "Database editor",
  "affiliation": "St. Paul Pioneer Press",
  "summary": "The St. Paul Pioneer Press used multiple regression analysis to identify schools in low-income areas that were outperforming their predicted standardized test scores.",
  "keywords": [
    "education data",
    "regression analysis",
    "test scores"
  ],
  "topics": [
    "Education",
    "Data & Databases",
    "Investigative Methods"
  ],
  "technologies": [
    "SPSS"
  ],
  "full_text": "The results from Minnesota's standardized test – used to measure progress for No Child Left Behind – yields a predictable list when sorted best to worst. At the top are schools from the wealthy neighborhoods and suburbs. At the bottom are schools from poor neighborhoods. This didn't seem like a fair way to assess how schools are performing. In June, the Pioneer Press published a series called 'Schools that Work,' stemming from a statistical analysis of test scores that revealed which elementary schools were doing better than expected, considering demographic factors. We found 13 schools serving predominantly low-income students that had consistently done better than expected. Then we visited those schools and came up with five things they were doing that experts said were keys to success. The bulk of the work was traditional street reporting, but data analysis pointed us in the right direction. Test score data is available in Minnesota from the state department of education. Numerous studies have shown that family income is the most significant predictor of test performance. The statistical procedure we used – regression – estimates the impact of income and uses that to 'predict' what each school's average should be. We used multiple regression with free/reduced-price lunch, English proficiency, and special education as independent variables. We used SPSS statistical software to run separate regressions on each set of scores and checked the 'residual,' which shows the difference between the predicted score and the actual score. We averaged the residuals to ensure schools labeled 'above predicted' were consistently better than expected. We decided to focus on schools with 60 percent or more low-income students to create a roadmap for success. We tapped other data sources like student demographics and teacher experience. Instead of publishing residuals, we identified schools as above, within, or below predicted to avoid confusion. I convinced my editors to put a good story in the paper; the analysis was just the path to get there.",
  "provenance": {
    "extracted_by": "pre-v2",
    "extracted_at": "",
    "verified": true
  },
  "similar": [
    {
      "slug": "1997-july-using-regression-analysis",
      "score": 0.416
    },
    {
      "slug": "1999-julyaugust-stats-and-such-raging-to-regress",
      "score": 0.278
    },
    {
      "slug": "1995-april-whats-affecting-test-scores",
      "score": 0.268
    },
    {
      "slug": "1998-august-car-cooking-raw-numbers",
      "score": 0.261
    },
    {
      "slug": "1996-september-count-me-in",
      "score": 0.254
    }
  ],
  "html_url": "https://dwillis.github.io/uplink/article/2006-july-august-finding-diamonds-in-the-rough.html",
  "json_url": "https://dwillis.github.io/uplink/article/2006-july-august-finding-diamonds-in-the-rough.json"
}