Education

How to assess a government and its education policies critically

The site gives no grades and no ranking of governments (ADR 0003). This page shows how to build a critical analysis with the data and the per-government pages, including adopted, kept and cancelled policies.

This visualization is descriptive and does not show that a government caused the change. See the methodology.

Five steps

1. Define “success” before looking at numbers. Access (enrollment), retention (dropout, age-grade lag), learning (Saeb, Pisa), equity (by income and race) and adult schooling are different things and do not always move together. Choose what will be measured, and why, before seeing which government “does well”.
2. Separate input, access and outcome. Spending and laws are inputs. Enrollment is access. Learning and completion are outcomes. The catalog has a lot of input and access and almost no outcomes: without learning data you cannot say quality improved.
3. Respect the lag. Mean years of schooling and adult illiteracy change mainly because younger, more schooled generations replace older ones: what changes in a term reflects decades of enrollment. Indicators for children and youth (net attendance, lag, completion) tend to respond sooner, but not necessarily in the same year. That is why the change table shows lags of zero to two years.
4. Compare with something. Compare with Latin America and the neighbors, which are in the table: if all rose together, the cause is probably not the government. And compare with the earlier trend: each page's “Pace before and during” section shows whether the series moved at the same pace before. Talk of change only when the pace broke.
5. To speak of impact, only with a causal design. Experiment, difference-in-differences, discontinuity or synthetic control, in a peer-reviewed study. Without that, the analysis is descriptive. Causal literature exists for few policies (such as Fundef and Bolsa Escola) and not for most.

Adopted, kept and cancelled policies

A policy has a longer history than the government that started it. In each “What started” card, the “Afterwards” block says what came next: kept, expanded, reduced, amended, replaced or ended, and by which norm.

When assessing an adopted policy, ask

• What problem did it target and in which metric did that problem show?
• Who runs it: the Union, states or municipalities? If the Union only induces, the result depends on who joined.
• When should the effect appear, and was the series already moving at that pace before?
• Was there an evaluation with a causal design? If not, treat any effect as a hypothesis.
• What else changed at the same time (income, demographics, other policies, the pandemic)?

When assessing a cancelled or replaced policy, ask

• Why did it end: negative evaluation, changed priorities, lack of funds or a merger with another?
• What replaced it, and was there a transition period or a gap?
• Did the original have evidence of effect, for or against? Cancelling something without effect and cancelling something that worked are different cases.
• The cost of switching: implementation time, lost teams and a broken data series.
• Who decided: the Executive, Congress or the courts?

When analyzing a cut, ask

• Is it appropriation, withholding (contingenciamento) or execution? The site shows execution; a block reversed within the year does not appear.
• In nominal or real terms? A figure that rises in current reais can fall in constant reais. The site's federal series are in constant reais.
• Is the spending the Union's or all levels'? The spending/GDP ratio rises when GDP falls, with no extra spending.
• Paid below committed can be deferral (restos a pagar), not a cut.
• Which metric could the cut reach, and how long would it take?

Common traps

• Crediting the president with what Congress, the courts, states or the economic cycle did. The per-government pages say when an act was Congress's.
• Years with a change of president: annual data cannot separate the two (the table flags these years).
• Methodology changes or series breaks: compare within the same series.
• Gross enrollment above one hundred percent (repeaters and over-age students): not an error, but needs care.
• Choosing the period, the comparison country or the indicator after seeing the result.

Where to see this on the site

Each per-government page gathers: what started (with each policy's fate), cuts and limits, change in the series, pace before and during, charts with neighbors, other factors and limitations. Some examples:

• Bolsonaro · 2019–2022
• Dilma · 2nd term (2015–2016)
• Lula · 3rd term (2023–)
• Temer · 2016 (acting during impeachment)

What the site does not measure yet

Learning (Saeb, Ideb, Pisa), functional illiteracy, authorized appropriation and withholding per year, enrollment by system and level before 2007 or 2012, and program-level series (Pronatec, Fies, ProUni, Pé-de-Meia). See the data audit. Education data audit.