Skip to content
Analyst work

Data studies

9 studies, each built around a decision a company or a school faces. Every page shows the data, the cleaning, the findings, my recommendation and where the evidence runs out.

  1. 01 · October 2026

    Duolingo streaks

    Do Duolingo’s streaks and leagues build language skill, or mostly daily logins?

    Headline finding
    Daily users grew 5.8x from Q4 2021 to Q2 2026. The median learner in Duolingo’s own study still needed 203 hours to reach half of B1, the intermediate level, which is 2.2 years of 15-minute days.
    My recommendation
    Reward mastery checks with the badges and league rank that XP earns today, and report a learning number beside daily users in every shareholder letter.
    Tools
    CSV tables with a source URL on every row · JavaScript on Node for the derived rates · D3.js for the first dashboard · React and SVG for the charts on this page
  2. 02 · October 2026

    Chegg after ChatGPT

    What should Chegg sell now that students have ChatGPT?

    Headline finding
    Chegg made $51.8 million in Q2 2026, 51% less than a year earlier. From 2022 to 2025, four of the five peers I compared it with grew revenue, led by Duolingo at 181%.
    My recommendation
    Run Chegg Study for cash and spend that cash on a guided tutor licensed to colleges. A sale of the company or its parts stays in reserve if no campus signs.
    Tools
    CSV tables with a source URL on every row · JavaScript on Node for the derived rates · D3.js for the first dashboard · React and SVG for the charts on this page
  3. 03 · October 2026

    TikTok STEM feed

    Why does TikTok’s US STEM feed trail Douyin in China, and what would make it work?

    Headline finding
    Douyin users each watched about 959 science videos in a year. TikTok’s STEM tab drew 0.84 views a year for each user it reached, less than one video each. The two counts cover different scopes. My reading is that where the feed sits and how teen limits are enforced drive the gap.
    My recommendation
    Make a capped teen mode the home screen for verified teens, and pay for more STEM videos to fill it. Speed up review and publish the minutes moved into STEM.
    Tools
    CSV tables with a source URL on every row · JavaScript on Node for the derived rates · D3.js for the first dashboard · React and SVG for the charts on this page
  4. 04 · October 2026

    Handshake and LinkedIn

    How can Handshake win new-grad hiring against LinkedIn and Indeed?

    Headline finding
    Recent graduates were out of work at 5.7% in June 2026, against 4.1% for all workers. Employers on Ashby’s hiring platform now sift 291 applications per hire. In NACE’s spring survey 43.4% of employers had detected AI-written applications and 20.1% used a tool to detect AI help.
    My recommendation
    Sell employers a verified applicant channel first. Then add alumni referrals and proof of skill from Handshake AI work on that verified base.
    Tools
    CSV tables with a source URL on every row · JavaScript on Node for the derived rates · D3.js for the first dashboard · React and SVG for the charts on this page
  5. 05 · October 2026

    AI and student learning

    Do coursework grades show what students remember, and what does an open AI chatbot do to the gap?

    Headline finding
    71.7% of the 4,649 students scored lower on the final exam than on their coursework. Students who studied in the most weeks of the term scored 5.9 points higher on the exam than the least consistent quarter, even after accounting for coursework grade and total online activity. Both figures are stored Open University results, and the code behind them was not kept.
    My recommendation
    The company should sell a hint-first tutor to schools, with weekly review sets built in. Schools should grade short weekly quizzes on old material and approve one guardrailed tutor.
    Tools
    JavaScript on Node for the derived rates · D3.js for the first dashboard · React and SVG for the charts on this page
  6. 06 · November 2026

    The case for third places

    Is there a real case for third places in Sacramento, and where should FourthSpace launch first?

    Headline finding
    Time with friends fell from 60.0 to 26.5 minutes a day between 2003 and 2025. Sacramento County is lonelier than 95% of U.S. counties, and 36 of its neighborhoods (census tracts) pair high loneliness with low walk access to any indoor place to gather.
    My recommendation
    Launch in south Sacramento, with the library branches near the loneliness hot spots as the first anchor venues.
    Tools
    Python 3.13 and pandas · statsmodels and SciPy for the models · Survey weights with Taylor-linearized variance · geopandas and networkx for the walk network · esda for the hot spots · matplotlib and Plotly for the first charts · JavaScript on Node for the numbers on this page · React and SVG for the charts on this page
  7. 07 · October 2026

    Oura and Whoop

    Does paying for a wearable change how people live?

    Headline finding
    From 2020 to 2024 the share of US adults using a wearable rose from 30.2% to 41.1%, and the share meeting the aerobic activity guideline went from 47.9% to 48.0%. Oura keeps 85% of paid members for a year. I found no figure in the coverage of its IPO filing for whether members sleep or move more.
    My recommendation
    Turn Advisor into a weekly plan and publish the share of members whose sleep or activity improved after 90 days.
    Tools
    CSV tables with a source URL on every row · JavaScript on Node for the derived rates · D3.js for the first dashboard · React and SVG for the charts on this page
  8. 08 · October 2026

    The protein boom

    Does the rush of high-protein products match a real gap in what people eat?

    Headline finding
    On average, no. Average intake in every age and sex group runs 1.25 to 1.74 times the recommended amount (the RDA). The shortfall sits with girls 14 to 18, where 20% fall below the requirement, and with adults 71 and older at 12% of men and 11% of women. Gen Z adults spend 2.8 times what Boomers spend on protein products.
    My recommendation
    Aim the new capacity at adults 65 and older and the family members who shop for them. Core Power keeps its athlete advertising at today’s budget.
    Tools
    CSV tables with a source URL on every row · JavaScript on Node for the derived rates · D3.js for the first dashboard · React and SVG for the charts on this page
  9. 09 · October 2026

    Campus rec and GPA

    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 method they share

  1. Start from a decision

    Each study names a client and the decision in front of them. I write the question so that its answer changes what they do.

  2. One row per figure

    Every figure sits in a row of a CSV file linked on its page, so a reader can open the row behind any number.

  3. One script per study

    A script reads the CSVs and works out every derived number. The page prints what the script returns, and a test fails if the two drift apart.

  4. Judgment is labeled

    Where I score something by hand, the table says so and lists the rows behind each score.

  5. Evidence behind the pick

    Each option I recommend cites the findings and rows behind it and names its main risk.

  6. Limits stay in

    Each page says what the data cannot support and what I would test with more access.