Adult Income Intelligence
How does the share of people earning above $50K differ across demographic and employment groups?
A four-page interactive Power BI report on the Adult Census Income dataset, built with Power Query and DAX. The key numbers were checked against a separate Pandas analysis of the same data.
Who earns above $50K?
The goal was to show how income differs by education, occupation, age, hours worked and sex, in a report anyone can filter and explore. The report needed to describe what the data shows clearly, without suggesting that any one factor causes higher income.
The Adult Census Income dataset
32,561 people from the 1994 U.S. Census, with 15 columns covering age, education, occupation, hours worked, marital status, sex and more. Income is a category, above or below $50K, not an exact salary.
The file had some gaps: 1,836 missing workclass values, 1,843 missing occupations and 583 blank countries. It also had 24 exact duplicate rows, which were kept, since identical records can belong to different people.
Cleaning in Power Query
- Set the column types so numbers behave like numbers and text sorts properly.
- Trim stray spaces so the same value is never counted as two different groups.
- Label missing values as "unknown" instead of deleting them. Dropping them would have removed 2,399 people (7.4% of the data).
- Add a 1 or 0 flag for earning above $50K, so counting high earners is a simple sum.
- Group age, weekly hours and education into readable bands, keeping the 16 original education levels for detail.
Rates, not counts, and insights without AI
Every comparison uses the share earning above $50K within each group. Counting high earners alone would favour large groups: there are 10,501 high school graduates but only 413 people with a doctorate.
The executive insight cards are written in DAX. Each one finds the group with the highest rate for the current filters, skips groups too small to be meaningful (under 20 people or 0.5% of the selection), and turns the result into a sentence. Change a filter and the insights rewrite themselves. No AI service is involved.
Observed results from the 1994 dataset
These describe this historical sample. They are observed differences, not causes, and they do not describe today’s workforce.
- Education: 48.5% of people with a bachelor’s degree or higher earned above $50K, compared with 13.0% of those with high school or less.
- Work: executive and managerial roles had the highest rate (48.4%), followed by professional roles (44.9%). The rate rose with hours worked up to 50 to 59 hours a week (44.6%).
- Age: the rate peaked at ages 45 to 54 (40.1%), and was 1.1% for people under 25.
- Sex: 30.6% of men and 10.9% of women earned above $50K.
The gap by sex appears at every education level. Among bachelor’s degree holders, 50.4% of men and 20.9% of women earned above $50K.
Income Patterns, and Education, Sex & Work. Click to enlarge.
Checked against Pandas
The key figures were recalculated in Python with Pandas on the same dataset and compared with the Power BI results. They match. This confirms the calculations agree; it does not test the assumptions behind the analysis.
| Check | Pandas | Power BI |
|---|---|---|
| People | 32,561 | 32,561 |
| Above $50K | 7,841 | 7,841 |
| Above $50K rate | 24.1% | 24.1% |
| Median weekly hours | 40 | 40 |
| Exact duplicate rows | 24 | 24 |
| Unknown workclass / occupation | 1,836 / 1,843 | 1,836 / 1,843 |
| Top education group | Doctorate, 74.1% | Doctorate, 74.1% |
| Top occupation | Exec-managerial, 48.4% | Exec-managerial, 48.4% |
The check also caught a detail worth handling carefully: the 583 unknown countries are blank cells rather than labelled values, so some Pandas versions count them as 0 unless they are converted first.
What this report can’t tell you
The data is from 1994 and does not describe today’s workforce. The results are observed differences: no other factors were controlled for, and education, occupation, age and sex overlap with each other. The census sampling weight was not used, so the results describe this sample rather than the whole U.S. population.
Looking at factors together
The natural next step is a model that accounts for several factors at once, to estimate how strongly each one is associated with income on its own rather than one at a time.
The GitHub repository includes the Power BI file (.pbix), the project source, the dataset and documentation.