Before asking AI to obtain data and draw a figure, specify the population and measurement. “Explore growth and unemployment” does not determine countries, periods, unemployment definitions, or whether to analyze levels or changes. Marx Max can help write and revise programs around a defined task; the researcher needs to inspect sources, actual code and output, and what the results support.
The accompanying World Bank six-country example develops this into a downloadable exercise. Fresh official API data were analyzed independently with Python and Stata in isolated environments and numerically compared. These are actual computations, not a measured instance of a model automatically completing the task.
Define an executable task
The example retains the United States, China, Japan, Germany, United Kingdom and Canada for 1991–2024. The year 1991 supplies the lag; plots and regressions use 1992–2024.
In a separate practice project, a request could be:
Use only public data from the official World Bank API. Obtain real annual GDP growth and total unemployment using ILO modeled estimates for the six countries, 1991–2024. Preserve raw responses, definitions, providers and retrieval times. Calculate annual unemployment changes within each country using consecutive years, without additional imputation. Plot complete observations for 1992–2024 in a 2-by-3 scatterplot, with separate linear fits and 95% confidence bands for the fitted means. Save Python code, Stata rerun code, CSV, figures and actual logs in this practice directory only. Describe associations without causal claims or country rankings.
This is an example natural-language request, not a fixed proprietary command. Execution depends on the current environment. For Stata, first follow the Stata guide to connect a licensed local installation. Save existing work and define reading and writing scope before continuing in an existing project.
Preserve sources separately from tidy data
The World Bank Indicators API supports public queries. This exercise selects World Development Indicators, source 2, with NY.GDP.MKTP.KD.ZG and SL.UEM.TOTL.ZS. The retrieval program restricts countries, years and indicators and checks pagination, duplicate keys and record counts.
Raw JSON preserves returned values and metadata; CSV reshapes the two series to country/year rows. Separate files help distinguish source revisions from preparation mistakes. provenance.json records request URLs, retrieval time, source revision dates and hashes. Website results refer to this snapshot; a future download may produce different values.
GDP growth is measured in annual %. The unemployment indicator measures unemployment as a share of the labor force and uses ILO modeled estimates, including source-provider imputations. Adding no imputation does not mean every source value is directly observed. The series does not unconditionally replace national headline statistics and should not be used for country rankings.
Check years, differences and actual samples
Unemployment change is this year’s rate minus last year’s rate, in percentage points, rather than percentage growth in the rate. Align consecutive years within countries and retain missing values; a difference across a gap is not an annual change.
This download contains 204 country/year rows without empty values in either indicator. Excluding 1991, which only supplies the lag, leaves 33 observations per country. Research should check planned years separately from the observations actually used, rather than judging completeness from an attractive chart.
What do the figure and regression answer?
Each country receives a separate linear regression of annual unemployment change on contemporaneous GDP growth with an intercept. Each point is one year. Bands are 95% confidence intervals for the fitted conditional mean, not prediction ranges for future years. Countries share axis scales, without pooling one common slope.
Negative fitted slopes are more pronounced for the United States and Canada; China’s fitted line is close to horizontal. Inspect data before deciding how they relate to an Okun relationship. The exercise uses conventional OLS standard errors without adjustments for annual serial correlation, pandemic outliers or structural change. Figures neither explain the causes of growth nor identify its causal effect on unemployment. Research needs a separately justified period, measurement and inference specification.
Rerun the snapshot before choosing to refresh it
The complete package includes raw API responses, input CSV, retrieval code, Python and Stata programs, both sets of figures and estimates, README and attribution. Both official indicator metadata pages state CC BY 4.0. Retain attribution and the dataset terms when reusing or redistributing data.
Extract into a fresh empty directory, retain a reference copy and follow the README from the package root. Installing Python dependencies accesses a software repository; analysis itself is offline. Stata independently reads the same CSV without Python or third-party Stata commands. Check log completion markers, six samples of 33 observations, coefficients and intervals. Both implementations were executed independently and compared on macOS; Windows hardware reruns are not covered.
Only explicitly run the retrieval program if you want fresh API data, then rerun analyses. Revisions can change results, so preserve the original snapshot and compare retrieval times. Programs overwrite named outputs in the practice directory; Stata clear all clears session memory. Save existing analyses first.
In Marx Max, use this public package in a separate practice project and continue from the actual code and outputs. Public data do not require a personal library. Cloud AI still receives conversation and tool-returned content; see research data processing. For your own research, establish data permissions and specifications again rather than carrying over explanations after changing country names.