# Master's and bachelor's earnings Lecture 11 loads the prepared `masters_earnings.csv` table. PS6 walks students through its construction from `masters_acs.csv`, a 335,109-row extract of the national ACS computer/math occupation family (SOCP starts with `15`). File preparation keeps the original columns listed below and adds the survey year. All ages, degrees, employment statuses, work schedules, and wage values in that occupation family are retained. Integer-valued Census fields are saved as integers to keep the download compact. There is no sampling or recoding of the source values. ```text Column Meaning year Survey year, added from the source filename AGEP Age in years SCHL Education code: 21 bachelor's; 22 master's ESR Employment status: 1–2 employed civilians COW Worker type: 1–5 civilian wage/salary workers WKHP Usual weekly hours worked WKWN Weeks worked in the previous year WAGP Wage and salary income in dollars ADJINC Census income-adjustment factor, with six implied decimal places PWGTP Person survey weight ``` On PS6, students inspect the codes and construct `eligible` and `earnings` with R pipelines. `masters_cpi.csv` contains the five survey years' annual CPI-U values from the existing price-index file. The homework joins it by year, labels the degree, converts annual pay to thousands of 2024 dollars, and selects the final columns. The lecture loads the prepared CSV and preloads completed analysis objects so later cells work independently. `masters_earnings.csv` contains 123,495 eligible person records from the national 2019, 2021, 2022, 2023, and 2024 American Community Survey one-year public-use files. Each year is a separate survey. All eligible records and their released wage values are retained. ```text Column Meaning year Survey year degree Bachelor's (SCHL 21) or Master's (SCHL 22) age Age in years pay Annual wage and salary income, in thousands of 2024 dollars weight ACS person survey weight (PWGTP) ``` `WAGP` measures wage and salary income over the previous 12 months. The sample contains currently employed civilian wage/salary workers (ESR 1-2, COW 1-5), ages 25-50, in computer and mathematical occupations (SOCP begins with `15`), with a bachelor's or master's degree. Workers report at least 35 usual weekly hours (WKHP) and at least 50 weeks worked (WKWN); wages and person weights must be positive. All five source files supply actual weeks worked in WKWN. The income calculation is: ```text pay = WAGP * ADJINC / 1000000 * CPI[2024] / CPI[year] / 1000 ``` ADJINC converts reported income to its survey year's dollars. Annual CPI-U means (BLS series CUUR0000SA0) then convert every year to 2024 purchasing power. The input is [cpi_annual.csv](../../../course_design/skill_returns/acs/cpi_annual.csv). Its SHA256 is `9271b90302db38bad1fb189b313742ccbc1f9095eaf596c5a6ec6e99e6eeca76`. The lecture's time-trend models use all 123,495 individual earnings records, giving each respondent equal weight and defining `time = year - 2022`. Annual averages are used for plotting and the observed pay-gap analysis. The retained survey weights support a separate instructor check. ```text Year BA sample MA sample BA mean MA mean Equal-weight gap% Survey-weighted gap% 2019 14,125 6,947 126.682 153.060 20.82 19.73 2021 15,568 7,763 132.480 161.193 21.67 21.11 2022 16,979 8,544 129.924 157.447 21.18 19.87 2023 17,707 8,834 129.094 158.843 23.04 22.59 2024 17,903 9,125 131.524 162.043 23.20 21.46 ``` Means are thousands of 2024 dollars. The gap is `100 * (master's mean / bachelor's mean - 1)`. Both summaries show a modestly larger mean-income gap in 2024 than in 2019. These records describe earnings among the sampled workers. They cannot establish the earnings gain caused by earning a master's degree or a change caused by AI. The previous-12-month income measure also spans calendar years. ## Lecture calculations Q1 fits a line to each degree group's individual earnings records, then uses those separate models to forecast earnings for 2026–2046. For 2028, the bachelor's forecast is $133,692.69 and the master's is $167,114.07. In the 2026 career scenario, the master's pays two resident annual fees of $28,886.38 and starts earning in 2029. The lecture sums these annual amounts; the first year-end when the master's cumulative total exceeds the bachelor's is 2037. This is a scenario based on continuing group earnings trends, not an estimate of an individual graduate's career progression. The two-slide PhD extension reads doctoral-degree records (`SCHL = 24`) from `masters_acs.csv`, using the same age and employment filters and CPI adjustment. It retains 4,427 records, including 1,063 in 2024, and fits `pay ~ time` to individual records. The scenario starts directly after the bachelor's in 2026, receives an assumed $40,000 annual stipend in 2024 dollars for five years with waived tuition, then earns forecast pay from 2032. Its cumulative earnings first reach the bachelor's total in 2038, seven years after completing the PhD. The existing bachelor's and master's paths are unchanged. The five-year duration and funding arrangement come from the [U-M Statistics PhD FAQ](https://lsa.umich.edu/stats/phd-students/ProspectiveGraduateStudents/prospective-phd-student-FAQs.html); the stipend amount is an example assumption. The plot is calculated in native R during the lecture build; the raw extract stays out of the bundle. Q2 tracks the observed annual degree pay gap around ChatGPT's November 2022 release, then tests whether the gap is growing across 2019–2024. The gap grows from $27,523 in 2022 to $30,519 in 2024. Combining the two lines with a degree indicator and its interaction with time expresses the slope difference as one coefficient. The interaction estimate is $782.25/year; the usual two-sided coefficient test gives p=.030674. The null is equal slopes. The positive estimate and p-value support a growing gap under this model. The test covers all five survey years. The fixed-pay calculations documented below remain a reference for PS6. The preparation script prints mean and median incomes, survey-weighted checks, and these ordinary least-squares fits: ```text 2024 individual records: pay ~ masters Intercept 131.523604505 masters 30.519185191 Individual records: pay ~ time * masters Intercept 130.053128984 time 0.606593006 masters 28.727902311 time:masters 0.782247059 Separate fitted lines: Bachelor's: fitted pay = 130.053128984 + 0.606593006 * time Master's: fitted pay = 158.781031296 + 1.388840064 * time ``` `masters` is 1 for a master's degree and 0 for a bachelor's degree. The fitted degree gap grows by about $782 per year over these five observed years. ### Break-even scenario Lecture 11 budgets a two-year full-time master's using published U-M LSA/Rackham tuition and required fees. The residential data science master's uses this fee schedule. The [tuition table](masters_tuition.csv) supplies each academic year's annual fee for Michigan residents and nonresidents; its [source notes](masters_tuition_README.md) document the rates and calculation. The main example uses Michigan resident fees. All costs and earnings are in 2024 dollars. The budget is twice the annual full-time fee for the selected year, representing four terms at that year's rate. This is a price-based estimate; it does not sum two consecutive academic years' bills. Tuition accumulates over the study years, with no wage earnings during school. Annual salaries are held at each degree group's observed mean. Actual credit loads and aid vary. Using pay in thousands of 2024 dollars: ```text tuition = 2 * annual_fee investment_cost = tuition + 2 * bachelor_pay earnings_gap = master_pay - bachelor_pay years_after_masters = investment_cost / earnings_gap years_since_bachelors = 2 + years_after_masters ``` The 2024 Michigan resident budget is $57,772.76. Including forgone earnings gives an investment of $320,820, an annual gap of $30,519, and 10.512 years after master's graduation to catch up (12.512 years after the bachelor's). The nonresident tuition budget is $115,552.76, giving 12.405 years after the master's. One study year at the same annual fee halves both tuition and forgone earnings, giving 5.256 years for a Michigan resident. ```text Year Tuition budget ($1,000s) Years after master's 2019 59.431 11.858 2021 58.416 11.263 2022 56.223 11.484 2023 56.672 10.584 2024 57.773 10.512 ``` Each row joins that survey year's observed degree means to the annual fee for the academic year starting in the same calendar year. Year-specific fees are converted to 2024 dollars with the same CPI-U series as earnings. The budget holds each year's rate constant for two study years. Fees increased 19.27% in nominal dollars from 2019 to 2024, while CPI rose 22.70%, so resident fees declined 2.79% in real terms. Between 2022 and 2024, fees rose in real terms, but the larger earnings gap reduced break-even time. The scenario uses pretax wages and excludes discounting, financing, tax differences, and wage growth. Living costs are assumed equal for the two paths. It describes a hypothetical calculation using observed group means, not an estimated causal return to earning a degree. Keep these details in the source notes and oral discussion, as requested by the instructor. ## Sources and reproduction Prepared October 5, 2026 from existing local Census extracts. The source project verified 2024 as the latest downloadable one-year release on October 1, 2026. Its [README](../../../course_design/skill_returns/acs/README.md) documents the downloads and official definitions. Official resources: [ACS microdata access](https://www.census.gov/programs-surveys/acs/microdata/access.html) and [2024 PUMS documentation](https://www.census.gov/programs-surveys/acs/microdata/documentation/2024.html). The local inputs are `course_design/skill_returns/acs/data/acs_YEAR.parquet`. Each adjacent `acs_YEAR.json` stores the exact download URL, archive SHA256, archive members, person-record count, and retained fields. The downloaded national source-archive hashes are: ```text Year Archive SHA256 2019 18e4ece4cc24781c01e8046c2d5afbabeb1f15452ddec43f60fdf3b1f6e67b92 2021 e1e64f81718e7d478cf2bf4ad8d1ae1fa03c690f6c86afa3984d1a6bd38dc8f2 2022 687a97773d0fdfec6c624ea6bf48466f12059d3ee80283a2ad8604ea7a66231e 2023 98b6ecb14b4830d1f2b54c265a5bd997828415c2dab14e17369c40e98b78d9d4 2024 afdc6d90c6e2f0bab365ed32d95ba4c4d8ac651162f46ac7861295b2dc469894 ``` From `f26`, regenerate the CSV and print the analysis using the existing ACS project environment: ```sh uv run --project course_design/skill_returns/acs python course_design/prepare_masters.py ``` The script reads the existing extracts and CPI table, and writes the source extract, the five-row price table, and the completed earnings reference. The PS6 checks compare the homework-built table to this reference row by row. The completed earnings CSV has 123,495 records, 4,364,786 bytes, and SHA256 `c07674b9393d172b15dee035ef92c136b863575c764e2c4af33cf84ee59ef9d1`. Pay values are saved to nine decimal places in thousands of dollars.