Measuring Institutions with the Worldwide Governance Indicators

Claudius Gräbner-Radkowitsch

Measuring Institutions with the Worldwide Governance Indicators

measurement

indicators

institutions

R

visualization

tutorial

BA

MA

How the World Bank’s Worldwide Governance Indicators turn a contested concept into six numbers: what the aggregation procedure assumes, which conception of good governance the dimensions encode, how to get the data into R, and what a correlation between governance scores and development outcomes can and cannot establish.

Author

Claudius Gräbner-Radkowitsch

Published

September 25, 2026

Modified

September 25, 2026

1 Introduction

How do you measure something as abstract as “institutions”? The outcomes of institutional arrangements are observable enough — growth rates, protests, corruption scandals — but the quality of the arrangements themselves is not. And unlike life expectancy or GDP, there is no agreed unit. Even the definition of an institution differs across research traditions, which means that any number claiming to measure institutional quality has already taken a position in a theoretical dispute before the first observation is collected (more details in Hodgson 2019).

The Worldwide Governance Indicators (WGI), produced by the World Bank since 1996, are the most influential attempt to produce such a number anyway. They cover more than 200 countries and territories along six dimensions, and they are used in aid allocation, investment risk assessment, and a very large empirical literature on institutions and development.

This page treats the WGI as a worked example of the general problem set out in The Theory-Ladenness of Observation: a contested concept, a chain of choices from concept to number, and a result that looks objective precisely because those choices are no longer visible in the final score. The aim is not to reject the indicators — they carry real information — but to use them knowing what they encode.

What you need

Basic R and the tidyverse. The data-handling patterns used here — caching a download, filtering out regional aggregates, joining on ISO-3 codes — are explained in Accessing Development Data in R.

Packages and the caching helper

library(WDI)
library(dplyr)
library(tidyr)
library(purrr)
library(readr)
library(tibble)
library(ggplot2)
library(ggpubr)
library(kableExtra)

wdi_cached <- function(indicators, file, ...) {
  path <- file.path("data", file)
  if (!file.exists(path)) {
    message("Cache miss - downloading from the World Bank API ...")
    write_csv(WDI(indicator = indicators, ...), path)
  }
  out <- read_csv(path, show_col_types = FALSE)
  # Guard against a retired indicator code (which returns silently without its
  # column) and against a cache file written by a different query
  stopifnot(all(names(indicators) %in% names(out)))
  out
}

2 The WGI Framework

2.1 The six dimensions

The WGI report institutional quality along six dimensions, which the World Bank groups into three pairs:

Table 1: The six WGI dimensions and their abbreviations

	Dimension	What it is meant to capture
Process by which governments are selected and replaced	Voice and Accountability (`va`)	Political participation, freedom of expression, media independence
	Political Stability and Absence of Violence (`pv`)	Likelihood of destabilisation or politically motivated violence
Capacity to formulate and implement policy	Government Effectiveness (`ge`)	Quality of public services, bureaucracy, policy implementation
	Regulatory Quality (`rq`)	Ability to formulate and implement sound private-sector policies
Respect for the institutions governing interaction	Rule of Law (`rl`)	Confidence in legal rules, property rights, police, and courts
	Control of Corruption (`cc`)	Extent to which public power is not exercised for private gain

2.2 How the scores are built

None of these dimensions is measured directly. Each is a composite built from dozens of underlying sources, of four broad kinds:

Expert assessments from think tanks, risk-rating firms, and NGOs
Firm surveys on the regulatory environment and on bribery
Household surveys on public services and on experience of corruption
Other indices, such as press-freedom rankings

These are combined using an Unobserved Components Model (Kaufmann et al. 2011). The model assumes that every source is a noisy signal of the same underlying, unobserved governance concept, and estimates that concept as a weighted average in which sources that agree more with the others receive more weight. The output is a score on an approximate \([-2.5, +2.5]\) scale, standardised so that the world mean is zero in every year, together with a standard error.

The scale is relative, not absolute

Because the distribution is re-standardised each year, a country whose score stays at 0.4 has not “stayed the same” — it has kept its position relative to the rest of the world. The WGI are therefore poorly suited to claims about global trends in governance, and comparisons over long time spans need care.

🎓 MA / PhD

The Unobserved Components Model earns its weights from inter-source agreement, which is exactly what makes the exercise circular when the sources are not independent. Many underlying assessments are produced by analysts who read each other’s reports and who know a country’s reputation, its credit rating, and its income level. Correlated measurement error of this kind is indistinguishable, inside the model, from genuine agreement about an underlying truth — so it inflates the apparent precision of the estimate.

Langbein and Knack (2010) push the point further and ask whether the six dimensions are separable at all. They find that the data are consistent with a single underlying factor, and argue that the six-dimension presentation implies a structure the statistics do not support. Thomas (2010) raises the prior question of construct validity: the model tells you how consistent the sources are with each other, not whether any of them measures “governance” — for that one needs an argument external to the data. Arndt and Oman (2006) add that the year-to-year changes in a country’s score are frequently smaller than the confidence interval around them, which is a problem for any use that rewards or penalises countries for movements in their rank.

3 Whose Governance? The Theory Behind the Index

3.1 Two traditions in institutional economics

To see what the WGI commit to, it helps to have two contrasting accounts of what institutions are in view. Both are live traditions in economics, and they ask different questions.

New Institutional Economics (NIE) treats institutions as “the rules of the game in a society” — the humanly devised constraints that shape interaction (North 1990). Its central question is how those rules reduce transaction costs and make exchange possible: secure property rights, enforceable contracts, predictable regulation, low expropriation risk. Institutions are assessed functionally, by how efficiently they support market exchange, and better institutions are expected to produce more investment and faster growth.

Original (or Old) Institutional Economics (OIE), in the tradition running from Veblen and Commons to Hodgson (2006), treats institutions as socially embedded systems of rules, habits, and power relations that constitute economic behaviour rather than merely constraining it. Its central questions are about power and distribution: whose interests a given rule serves, how institutions are reproduced and contested, and how they change. On this view there is no institutional arrangement that is efficient in general — only arrangements that are advantageous to some parties, in some context, for some purpose.

The difference matters here because an indicator has to pick one. A measure built on NIE premises will score highly a state that enforces contracts cheaply; a measure built on OIE premises would want to know who wrote the contracts.

3.2 Exercise: mapping the dimensions

Exercise

Work through Table 1 one dimension at a time and fill in the table below. For each dimension, ask: what would an NIE account emphasise here? What would an OIE account want to know that the dimension does not report? And what institutional feature does the dimension leave out altogether?

WGI dimension	NIE focus (property rights, contracts, transaction costs)	OIE concerns (power, embeddedness, distribution)	Missing aspects
Rule of Law
Regulatory Quality
Government Effectiveness
Control of Corruption
Voice and Accountability
Political Stability

Possible solution

WGI dimension	NIE focus	OIE concerns	Missing aspects
Rule of Law	Strong focus on protection of property rights and contract enforcement	Little attention to whose interests the law serves, or to power asymmetries in access to it	Labour rights, collective bargaining institutions
Regulatory Quality	Emphasis on private-sector-friendly policy and low business cost	Distributional effects invisible; market solutions treated as the benchmark	Industrial policy, state capacity for structural transformation
Government Effectiveness	Efficient service delivery and bureaucratic quality	Democratic input into what services are delivered is not part of the score	Welfare-state institutions, participatory administration
Control of Corruption	Rent-seeking as a transaction cost to be minimised	Focus on individual transgression rather than systemic capture	Lobbying, campaign finance, revolving doors — legal influence
Voice and Accountability	Mainly procedural: elections, free press	Some attention to participation, though thinly	Economic democracy, worker voice, political inequality
Political Stability	Predictable environment for investment	Status-quo bias: legitimate grievance and disruptive reform both lower the score	Social movements, transformative institutional change

The pattern. The WGI track NIE concerns closely and OIE concerns barely. This is not a flaw in the statistics; it is the concept the index was built to measure. The framework grew out of the “good governance” agenda of the 1990s, whose reform template Chang (2011) calls the Global Standard Institutions — a specific set of Anglo-American arrangements presented as universally appropriate, and, he argues, not the set that today’s rich countries actually used while they were developing.

3.3 What the index encodes, and what it omits

The exercise above makes four commitments visible:

A market-friendly benchmark. Regulatory Quality rewards policy that lowers costs to private business. Regulation protecting workers, consumers, or ecosystems enters, if at all, as a cost.
A liberal-democratic template. Voice and Accountability is built around Western electoral and media institutions, and reads other forms of participation poorly.
A stability bias. Political Stability penalises disruption as such, including disruption that is a response to injustice, which can flatter orderly authoritarian regimes.
Transgression rather than structure. Control of Corruption is oriented to identifiable illegal acts, and largely misses the legal channels through which concentrated interests shape policy.

And a corresponding set of institutional domains receives little weight anywhere in the index: labour relations and social insurance, industrial and innovation policy, environmental governance, and the formal and informal rules governing gender relations.

None of this makes the WGI useless. It makes them an indicator of something in particular, and the methods-section question is whether that something is what your research question is about.

4 Getting the Data into R

4.1 Downloading the estimates

The WGI live in their own World Bank database (source 3), not in the World Development Indicators. That means the WDI() function will not find them — WDI() queries source 2 — and the indicator codes are prefixed, e.g. GOV_WGI_CC.EST for the Control of Corruption estimate. Each dimension also publishes a percentile rank (.SC), a standard error (.SE), the number of sources (.SR), and the bounds of a 90% confidence interval (.SC_LB, .SC_UB).

Downloading the WGI from the World Bank API

library(jsonlite)
options(timeout = 300)

wgi_codes <- c(va = "VA", pv = "PV", ge = "GE", rq = "RQ", rl = "RL", cc = "CC")

fetch_wgi <- function(code, start, end) {
  url <- paste0(
    "https://api.worldbank.org/v2/country/all/indicator/GOV_WGI_", code, ".EST",
    "?source=3&format=json&per_page=20000&date=", start, ":", end
  )
  raw <- fromJSON(url)[[2]]
  tibble(
    iso3c = raw$countryiso3code,
    year  = as.integer(raw$date),
    value = raw$value
  ) |>
    # Nine entities (Anguilla, Jersey, Taiwan, …) are returned without an ISO-3
    # code and would collide with each other when the data are pivoted wide
    filter(iso3c != "")
}

wgi_long <- imap_dfr(wgi_codes, \(code, nm) {
  fetch_wgi(code, 2000, 2024) |> mutate(indicator = nm)
})

# The API returns regional and income aggregates alongside countries; WDI's
# country table is the cleanest way to drop them
countries <- WDI::WDI_data$country |>
  as_tibble() |>
  filter(region != "Aggregates") |>
  select(iso3c, country, region, income)

wgi_long |>
  pivot_wider(names_from = indicator, values_from = value) |>
  inner_join(countries, by = "iso3c") |>
  select(country, iso3c, region, income, year, va, pv, ge, rq, rl, cc) |>
  filter(if_any(va:cc, \(x) !is.na(x))) |>
  arrange(country, year) |>
  write_csv("data/wgi-estimates.csv")

Code

wgi <- read_csv("data/wgi-estimates.csv", show_col_types = FALSE) |>
  mutate(governance_avg = rowMeans(pick(va:cc), na.rm = TRUE))

dimension_labels <- c(
  va = "Voice & Accountability",
  pv = "Political Stability",
  ge = "Government Effectiveness",
  rq = "Regulatory Quality",
  rl = "Rule of Law",
  cc = "Control of Corruption"
)

Code

wgi |>
  filter(year == 2022) |>
  slice_head(n = 6) |>
  select(country, iso3c, year, va, pv, ge, rq, rl, cc, governance_avg) |>
  kable(digits = 2) |>
  kable_styling(bootstrap_options = c("striped", "hover"))

Table 2: Structure of the cached WGI file: one row per country and year, one column per dimension.

country	iso3c	year	va	pv	ge	rq	rl	cc	governance_avg
Afghanistan	AFG	2022	-1.84	-2.45	-1.96	-1.42	-1.86	-1.17	-1.78
Albania	ALB	2022	0.02	0.01	0.19	0.02	-0.13	-0.50	-0.06
Algeria	DZA	2022	-1.07	-0.81	-0.27	-0.88	-0.80	-0.64	-0.74
American Samoa	ASM	2022	0.94	1.25	0.77	0.39	1.19	1.18	0.95
Andorra	AND	2022	0.95	1.55	1.16	1.10	1.41	1.18	1.22
Angola	AGO	2022	-1.05	-0.54	-0.83	-0.86	-1.16	-0.60	-0.84

The file covers 207 countries. Note that the years are not consecutive: the WGI were produced biennially until 2002 and annually since.

governance_avg is a convenience, not an official indicator

The unweighted mean of the six dimensions used throughout this page is a shortcut for exposition. The World Bank does not publish it, and averaging six scores that are already composites compounds the aggregation problem rather than solving it. For actual work, report the dimension your argument is about.

4.2 Working with the uncertainty measures

The estimate is only half of what the WGI publish. Each score comes with a standard error (.SE), and each percentile rank with the bounds of a 90% confidence interval. These are not decoration: for many country pairs, the intervals overlap so heavily that the ranking between them carries no information at all.

This page uses the estimates alone to keep the code readable. If you use the WGI in your own work, download the .SE series alongside .EST — the download function above needs only a different suffix — and check whether the differences you want to interpret survive their confidence intervals.

5 Exploring the Indicators

5.1 Trends over time

Code

selected <- c("DNK", "USA", "BRA", "IND", "NGA")

wgi |>
  filter(iso3c %in% selected) |>
  pivot_longer(va:cc, names_to = "indicator", values_to = "score") |>
  # Keep the World Bank's own ordering of the dimensions, not alphabetical
  mutate(indicator = factor(dimension_labels[indicator],
                            levels = dimension_labels)) |>
  ggplot(aes(x = year, y = score, colour = country)) +
  geom_line(linewidth = 0.8) +
  facet_wrap(~indicator) +
  scale_x_continuous(breaks = c(2000, 2010, 2020)) +
  scale_colour_viridis_d(option = "H") +
  labs(
    x = NULL,
    y = "WGI score (approx. −2.5 to 2.5)",
    colour = NULL
  ) +
  theme_minimal() +
  theme(legend.position = "bottom")

Figure 1: WGI dimensions for five countries, 2000–2024. Remember that the scale is re-standardised each year, so a flat line means a constant position relative to the world, not constant institutional quality.

5.2 How much independent information do six dimensions carry?

Table 3 reports the correlations between the six dimensions across countries, using each country’s average over the whole period.

Between-country and within-country correlation matrices

# Between: one observation per country, so the variation is cross-sectional
wgi_country_means <- wgi |>
  summarise(across(va:cc, \(x) mean(x, na.rm = TRUE)), .by = c(country, iso3c))

cor_between <- cor(wgi_country_means |> select(va:cc), use = "complete.obs")

# Within: subtract each country's own mean, leaving only movement over time.
# This is the same transformation a fixed-effects model applies.
wgi_within <- wgi |>
  filter(if_all(va:cc, \(x) !is.na(x))) |>
  mutate(across(va:cc, \(x) x - mean(x, na.rm = TRUE)), .by = iso3c)

cor_within <- cor(wgi_within |> select(va:cc), use = "complete.obs")

Code

cor_between |>
  as.data.frame() |>
  rownames_to_column("Dimension") |>
  mutate(Dimension = dimension_labels[Dimension]) |>
  kable(digits = 2, col.names = c("Dimension", "VA", "PV", "GE", "RQ", "RL", "CC")) |>
  kable_styling(bootstrap_options = c("striped", "hover"))

Table 3: Correlations between the WGI dimensions across countries (country means, 2000–2024).

Dimension	VA	PV	GE	RQ	RL	CC
Voice & Accountability	1.00	0.78	0.82	0.82	0.91	0.85
Political Stability	0.78	1.00	0.77	0.72	0.86	0.82
Government Effectiveness	0.82	0.77	1.00	0.96	0.93	0.95
Regulatory Quality	0.82	0.72	0.96	1.00	0.91	0.91
Rule of Law	0.91	0.86	0.93	0.91	1.00	0.96
Control of Corruption	0.85	0.82	0.95	0.91	0.96	1.00

Government Effectiveness, Regulatory Quality, Rule of Law, and Control of Corruption move together very closely across countries — every pairwise correlation in that block is above 0.9. Whatever the six dimensions separate conceptually, four of them are close to interchangeable empirically, which is the statistical version of Langbein and Knack (2010)’s argument and a good reason not to put two of them on the right-hand side of the same regression.

5.3 Between countries versus within countries

A correlation computed on country means answers a between-country question: do countries that score well on one dimension score well on the others? A different question is whether the dimensions move together within a country over time. The two can differ sharply, and conflating them is a common mistake.

To separate them, subtract each country’s own mean from its observations and correlate the deviations — the wgi_within object above. What remains is the within-country variation, which is exactly what a fixed-effects model uses; see Panel Data and Fixed Effects in R.

Code

mean_offdiag <- function(m) mean(m[upper.tri(m)])
min_offdiag  <- function(m) min(m[upper.tri(m)])
max_offdiag  <- function(m) max(m[upper.tri(m)])

tibble(
  Comparison = c(
    "Between countries (country means)",
    "Within countries (deviations from country mean)"
  ),
  `Mean correlation` = c(mean_offdiag(cor_between), mean_offdiag(cor_within)),
  `Lowest pair`      = c(min_offdiag(cor_between), min_offdiag(cor_within)),
  `Highest pair`     = c(max_offdiag(cor_between), max_offdiag(cor_within))
) |>
  kable(digits = 2) |>
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Table 4: Average correlation between the six dimensions, computed between countries and within countries over time.

Comparison	Mean correlation	Lowest pair	Highest pair
Between countries (country means)	0.86	0.72	0.96
Within countries (deviations from country mean)	0.40	0.22	0.58

Code

as_long <- function(m, label) {
  m |>
    as.data.frame() |>
    rownames_to_column("var1") |>
    pivot_longer(-var1, names_to = "var2", values_to = "correlation") |>
    mutate(
      var1  = factor(var1, levels = names(dimension_labels)),
      var2  = factor(var2, levels = names(dimension_labels)),
      panel = label
    ) |>
    filter(as.integer(var1) < as.integer(var2))
}

bind_rows(
  as_long(cor_between, "Between countries"),
  as_long(cor_within, "Within countries")
) |>
  mutate(panel = factor(panel, levels = c("Between countries", "Within countries"))) |>
  ggplot(aes(x = var1, y = var2, fill = correlation)) +
  geom_tile(colour = "white", linewidth = 0.5) +
  geom_text(aes(label = round(correlation, 2)), size = 3.2) +
  facet_wrap(~panel) +
  scale_fill_gradient(low = "#f7f7f7", high = "#e08214", limits = c(0, 1)) +
  # toupper() as a function maps each level to its own label; a positional
  # vector would misalign, since the dropped diagonal leaves five levels per axis
  scale_x_discrete(labels = toupper) +
  scale_y_discrete(labels = toupper) +
  labs(x = NULL, y = NULL, fill = "Correlation") +
  theme_minimal() +
  theme(panel.grid = element_blank(), axis.ticks = element_blank()) +
  coord_fixed()

Figure 2: The two correlation matrices side by side, on a shared scale. Between countries (left) every pair is above 0.7; within countries (right) the same pairs fall to between 0.2 and 0.6.

The between-country correlations are roughly twice the within-country ones. The near-collinearity of the WGI is therefore a cross-sectional phenomenon: rich, stable countries score well on everything and poor, unstable ones score badly on everything, and that single contrast dominates the variance. Once you look at how a given country’s scores change over time, the dimensions come apart considerably. Voice and Accountability and Government Effectiveness correlate at 0.82 between countries but only 0.22 within them: countries do lose ground on political voice while their bureaucracies hold up, and the reverse.

Why this distinction is worth the trouble

If you run a cross-section, the WGI will behave as one variable no matter how many dimensions you enter. If you exploit within-country variation, they carry more independent information — but the year-to-year movements are also where the measurement error is largest relative to the signal, so more of what you are modelling may be noise. Each design comes with costs.

6 Governance and Development Outcomes

6.1 The bivariate picture

The empirical claim that made institutional quality a central variable in development economics is that good institutions cause growth (Acemoglu et al. 2001). The raw correlation is easy to reproduce. We pull three outcomes from the WDI, average them over 2018–2022 to match the governance data, and join on ISO-3 codes.

Code

outcomes <- wdi_cached(
  indicators = c(
    gdp_pc_ppp = "NY.GDP.PCAP.PP.CD",
    gini       = "SI.POV.GINI",
    life_exp   = "SP.DYN.LE00.IN"
  ),
  file  = "wdi-governance-outcomes.csv",
  start = 2018,
  end   = 2022
) |>
  summarise(
    across(c(gdp_pc_ppp, gini, life_exp), \(x) mean(x, na.rm = TRUE)),
    .by = iso3c
  )

governance_outcomes <- wgi |>
  filter(year >= 2018, year <= 2022) |>
  summarise(governance_avg = mean(governance_avg, na.rm = TRUE),
            .by = c(country, iso3c, region)) |>
  inner_join(outcomes, by = "iso3c")

Joining on iso3c rather than on the country name is not a stylistic preference. The WGI call the country Egypt, Arab Rep. in some vintages and Egypt in othersm, while the WDI use their own spellings. A name join silently drops the rows that disagree, and produces a silently shorter dataset is a common error source, and you should be aware of it.

Code

plot_outcome <- function(data, yvar, ylab) {
  ggplot(data, aes(x = governance_avg, y = {{ yvar }})) +
    geom_point(alpha = 0.6, size = 1.8) +
    geom_smooth(method = "lm", formula = y ~ x, se = TRUE) +
    labs(x = "Average WGI score", y = ylab) +
    theme_minimal()
}

ggarrange(
  plot_outcome(governance_outcomes, log(gdp_pc_ppp), "GDP per capita (PPP, log)"),
  plot_outcome(governance_outcomes, life_exp, "Life expectancy (years)"),
  plot_outcome(governance_outcomes, gini, "Inequality (Gini)"),
  ncol = 3, nrow = 1
)

Figure 3: Average WGI score against three development outcomes, 2018–2022 averages. Each point is a country.

Governance scores correlate strongly and positively with income and life expectancy. The association with the Gini coefficient is negative — higher governance scores go with lower measured inequality — but it is much weaker and much noisier than the other two, and the Gini is available for far fewer countries. That weakness is itself informative: an index built around contract enforcement and regulatory predictability has no particular reason to track distribution, and it does not.

6.2 What the correlation cannot tell you

Figure 3 is consistent with at least four stories. But the info from the figure is, unfortunately, unable to distinguishes them:

Institutions cause development — the NIE claim, and the reading Acemoglu et al. (2001) defend with an instrumental-variables strategy precisely because the raw correlation cannot settle it.
Development causes institutions — rich states can afford professional bureaucracies, well-paid judges, and functioning statistical offices.
Something else causes both — colonial history, geography, or the structure of the export basket, which is the mechanism explored in Income, Output, and Complexity Indicators.
The measurement is partly circular — assessors who rate governance know how rich a country is. Kaufmann et al. (2011) acknowledge the risk; Arndt and Oman (2006) argue it is substantial. To the extent that perceptions of institutional quality are formed partly by observed prosperity, part of this correlation is built into the indicator rather than discovered by it.

Story 4 is the one specific to indicator-based work, and the one most often ignored. Distinguishing the first three requires an identification strategy — see Reasoning with DAGs for how to state what your design assumes, and Panel Data and Fixed Effects in R for the most common way of exploiting within-country variation.

7 How the WGI Are Used

Knowing where these numbers end up is part of knowing what they are.

Aid allocation and eligibility. The United States’ Millennium Challenge Corporation uses Control of Corruption and Rule of Law among its eligibility indicators, so the scores have a direct budgetary consequence. Multilateral development banks use governance ratings in country allocation formulas, and several donors report against them.

Investment and credit risk. Sovereign risk models and country-risk products draw on the WGI directly. This creates a feedback loop worth naming: an index partly built from expert perceptions feeds into ratings that shape those same experts’ views of a country.

Academic research. The WGI are among the most frequently used variables in cross-country empirical work, typically as a control for institutional quality or as the outcome of interest. Table 3 is the reason to be careful here: entering three or four dimensions as separate regressors imports severe multicollinearity, and picking whichever single dimension supports the hypothesis is a specification search.

Policy targeting. Governments do set targets in terms of WGI rank. Because the scores are re-standardised annually and the confidence intervals are wide, rank movements are frequently not statistically distinguishable from zero, which makes them poor targets — and, as Arndt and Oman (2006) note, creates an incentive to manage perceptions rather than institutions.

8 Summary

The WGI report six dimensions of governance, each a composite of dozens of expert, firm, and household assessments aggregated by an Unobserved Components Model that weights sources by their agreement with each other.
The scale is relative and re-standardised every year, so scores describe position in the world distribution, not an absolute level.
Across countries, four of the six dimensions are nearly interchangeable; within countries over time they carry considerably more independent information. Which fact applies to you depends on your research design.
The index encodes a specific, NIE-flavoured conception of good governance — property rights, contract enforcement, regulatory predictability — and gives little weight to labour, welfare, industrial-policy, environmental, and gender institutions.
Governance scores correlate strongly with income and life expectancy and weakly with inequality. The correlation with income is compatible with causation in either direction, with common causes, and with partial circularity in the measurement itself.
Use the uncertainty measures the WGI publish. A difference smaller than its confidence interval is not a finding.

9 Further Reading

Kaufmann et al. (2011) — the authors’ own account of the methodology and their reply to the main criticisms. Start here; it is more candid about the limitations than most secondary discussions suggest.
Langbein and Knack (2010) — the case that the six dimensions measure one thing.
Thomas (2010) — the construct-validity question: do the WGI measure what they claim to measure, and how would we know?
Arndt and Oman (2006) — a book-length treatment of how governance indicators are used and misused, with particular attention to aid allocation.
Chang (2011) — institutional quality from a development-policy and economic-history perspective, and the source of the “Global Standard Institutions” critique.
Hodgson (2006) — what an institution is, from the OIE side.
North (1990) — the founding statement of the NIE position.

References

Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2001. “The Colonial Origins of Comparative Development: An Empirical Investigation.” American Economic Review 91 (5): 1369–401. https://doi.org/10.1257/aer.91.5.1369.

Arndt, Christiane, and Charles Oman. 2006. Uses and Abuses of Governance Indicators. Development Centre Studies. OECD Publishing. https://doi.org/10.1787/9789264026865-en.

Chang, Ha-Joon. 2011. “Institutions and Economic Development: Theory, Policy and History.” Journal of Institutional Economics 7 (4): 473–98. https://doi.org/10.1017/S1744137410000378.

Hodgson, Geoffrey M. 2019. “Taxonomic Definitions in Social Science, with Firms, Markets and Institutions as Case Studies.” Journal of Institutional Economics 15 (2): 207–33. https://doi.org/10.1017/s1744137418000334.

Hodgson, Geoffrey M. 2006. “What Are Institutions?” Journal of Economic Issues 40 (1): 1–25. https://doi.org/10.1080/00213624.2006.11506879.

Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi. 2011. “The Worldwide Governance Indicators: Methodology and Analytical Issues.” Hague Journal on the Rule of Law 3 (2): 220–46. https://doi.org/10.1017/S1876404511200046.

Langbein, Laura, and Stephen Knack. 2010. “The Worldwide Governance Indicators: Six, One, or None?” The Journal of Development Studies 46 (2): 350–70. https://doi.org/10.1080/00220380902952399.

North, Douglass C. 1990. Institutions, Institutional Change and Economic Performance. Cambridge University Press. https://doi.org/10.1017/CBO9780511808678.

Thomas, Melissa A. 2010. “What Do the Worldwide Governance Indicators Measure?” The European Journal of Development Research 22 (1): 31–54. https://doi.org/10.1057/ejdr.2009.32.

Citation

BibTeX citation:

@inreference{graebner2026-measuring-institutions-wgi,
  author = {Gräbner-Radkowitsch, Claudius},
  editor = {Gräbner-Radkowitsch, Claudius},
  title = {Measuring {Institutions} with the {Worldwide} {Governance}
    {Indicators}},
  booktitle = {Method and Inquiry},
  date = {2026-09-25},
  url = {https://euf-methodology-hub.netlify.app/resources/measuring-institutions-wgi},
  langid = {en}
}

For attribution, please cite this work as:

Gräbner-Radkowitsch, Claudius. 2026. “Measuring Institutions with the Worldwide Governance Indicators.” In Method and Inquiry, edited by Claudius Gräbner-Radkowitsch. September 25. https://euf-methodology-hub.netlify.app/resources/measuring-institutions-wgi.