Skip to main content
v2026.11,986 entries · CC-BY 4.0

ESG Ratings: How They Are Built and Why Raters Disagree

How ESG ratings are built, why providers disagree, and how to compare two raters with a worked correlation example and code.

Written and maintained by CASRAI Editorial Board

Published

On this page

ESG ratings are scores or letter grades that a data provider assigns to a company to summarise how it handles environmental, social and governance issues. Two raters can score the same firm on completely different scales, and some scales run in opposite directions, so a rating is only usable once you know who built it, how, and on what date. This guide shows how two major providers construct their ratings, why raters disagree, and how to compare two raters with a correlation, including a fully worked example.

What are ESG ratings?

An ESG rating is a provider’s assessment of a company’s exposure to, and management of, environmental, social and governance issues. Ratings are produced by commercial data providers using their own methodologies; they are not issued by a regulator and are not an audited disclosure. For the difference between rating a company and the company reporting on itself, see the sustainability reporting guide, which covers frameworks and the EU CSRD timeline.

How do MSCI and Sustainalytics build their ratings?

The two providers answer different questions, which is why their outputs cannot be placed on one axis without a transformation. The table traces each statement to the provider’s own page (both read 2026-11-11).

Feature MSCI ESG Ratings Sustainalytics ESG Risk Ratings
Output An industry-relative letter rating from AAA to CCC One of five risk levels: negligible, low, medium, high, severe
What is measured Industry leaders and laggards, rated relative to peers, using issues material to each GICS sub-industry A company’s exposure to industry-specific material ESG risks and its management of those risks; an absolute measure of risk
Method Rules-based methodology Six steps: total exposure, manageable risk, unmanageable risk, managed risk, management gap, unmanaged risk
Better means Higher on the letter scale (AAA is the top) Lower risk (negligible is the best level)

Two consequences follow. First, MSCI letters are relative to industry peers while Sustainalytics categories describe absolute unmanaged risk, so the two are not measuring the same thing even before data differences enter. Second, the scales point in opposite directions: a high MSCI grade and a low Sustainalytics risk level are both favourable. Sources: MSCI ESG Ratings; Sustainalytics ESG data.

Why do ESG ratings from different providers disagree?

Berg, Kölbel and Rigobon (2022) compared six raters (KLD, Sustainalytics, Moody’s ESG, S&P Global, Refinitiv and MSCI) and found pairwise correlations between their ratings ranging from 0.38 to 0.71. They decomposed the divergence into three sources:

Source of divergence Meaning Approximate share
Measurement Raters measure the same attribute with different indicators about 56%
Scope Raters include different attributes in the rating about 38%
Weight Raters weight the same attributes differently about 6%

They also report a rater effect: a rater’s overall impression of a firm influences its scores on individual categories. The authors conclude that the results call for greater attention to how the data underlying ESG ratings are generated. In practice this means that the correlation between two raters is a benchmark you should expect to be moderate, not near 1.

Worked example: how correlated are two ESG raters?

Illustrative hypothetical data, not real company ratings. Suppose eight firms (A to H) are scored by two raters. Rater 1 uses an MSCI-style letter scale, coded AAA = 7, AA = 6, A = 5, BBB = 4, BB = 3, B = 2, CCC = 1. Rater 2 uses a Sustainalytics-style risk category, coded negligible = 1, low = 2, medium = 3, high = 4, severe = 5.

Step 1: put both scales on the same direction

Because lower risk is better, reverse-code Rater 2 as 6 − x, so that higher is better on both.

Firm Rater 1 letter Rater 1 code (x) Rater 2 risk code Reversed (y = 6 − code)
A AA 6 2 4
B A 5 2 4
C A 5 3 3
D BBB 4 3 3
E BB 3 4 2
F BBB 4 2 4
G B 2 5 1
H BB 3 3 3

Step 2: apply the Pearson formula

The Pearson correlation is r = Σ(x − x̄)(y − ȳ) / √[Σ(x − x̄)² × Σ(y − ȳ)²], where x̄ and ȳ are the means. The full formula is explained in Formula for Correlation Analysis: Pearson r Step by Step.

  • Means: x̄ = 32 / 8 = 4.0; ȳ = 24 / 8 = 3.0.
  • Deviations of x: 2, 1, 1, 0, −1, 0, −2, −1.
  • Deviations of y: 1, 1, 0, 0, −1, 1, −2, 0.
  • Σ(x − x̄)(y − ȳ) = 2 + 1 + 0 + 0 + 1 + 0 + 4 + 0 = 8.
  • Σ(x − x̄)² = 4 + 1 + 1 + 0 + 1 + 0 + 4 + 1 = 12; Σ(y − ȳ)² = 1 + 1 + 0 + 0 + 1 + 1 + 4 + 0 = 8.
  • r = 8 / √(12 × 8) = 8 / √96 = 8 / 9.798 = 0.816.

Skipping Step 1 gives r = −0.816 on the raw codes: the same agreement, with the wrong sign. This is the most common error when combining a rating with a risk score.

How to interpret the correlation between raters

Result Reading
r near +1 after reversing scales The raters rank firms almost identically in your sample
r = 0.82 in the example Strong agreement in this small hypothetical set; it is higher than the 0.38 to 0.71 range Berg et al. report for six real raters, so do not treat it as typical
r near 0 The ratings carry unrelated information about your sample
Negative r after reversing Check the coding direction first, then consider genuine disagreement

For general guidance on strength and direction, see Correlation Coefficient: What It Measures, Pearson vs. Spearman, and How to Report It.

Assumptions and common mistakes

  • Comparing letters with risk scores without reversing. The scales run in opposite directions, as shown above.
  • Treating ordinal grades as interval numbers. Letter grades and risk levels are ordered categories. Coding them 1 to 7 assumes equal spacing; a rank-based coefficient such as Spearman’s correlation is the safer choice when that is doubtful. See levels of measurement.
  • Mixing relative and absolute measures. MSCI ratings are relative to industry peers; Sustainalytics measures absolute risk. A correlation across industries mixes both effects.
  • Treating a rating as audited fact. A rating is a provider’s assessment built from its own methodology; it is a measurement that can contain measurement bias.
  • Mixing rating dates. Providers update ratings over time, so compare ratings taken on the same date and record that date.
  • Small samples. With eight firms, r is unstable; the example is for arithmetic, not inference.

How to report ESG ratings and their correlation (APA style)

Name the provider, the product, the rating scale, the retrieval date and the coding you applied. A correlation is reported in APA style as r(df) = value, p = value, with no leading zero for statistics bounded by 1:

Hypothetical result sentence: After reverse-coding the risk categories so that higher values indicated better performance, the two raters’ scores were strongly correlated, r(6) = .82, p = .013.

For the research literature, a reference entry for the paper cited here is: Berg, F., Kölbel, J. F., & Rigobon, R. (2022). Aggregate confusion: The divergence of ESG ratings. Review of Finance, 26(6), 1315–1344. https://academic.oup.com/rof/article/26/6/1315/6590532

How to calculate the correlation in Excel, R or Python

Each snippet uses the eight-firm example (x = Rater 1 code, y = reversed Rater 2 code) and returns 0.816.

Excel

Put x in B2:B9 and the raw risk codes in C2:C9. In D2 enter =6-C2 and fill down, then:

=CORREL(B2:B9, D2:D9)

R

x = c(6, 5, 5, 4, 3, 4, 2, 3)
risk = c(2, 2, 3, 3, 4, 2, 5, 3)
y = 6 - risk
cor(x, y)
cor.test(x, y)
cor(x, y, method = "spearman")

Python

from statistics import correlation  # Python 3.10+
x = [6, 5, 5, 4, 3, 4, 2, 3]
risk = [2, 2, 3, 3, 4, 2, 5, 3]
y = [6 - r for r in risk]
print(round(correlation(x, y), 3))  # 0.816

For a spreadsheet walkthrough, see How to Find the Correlation Coefficient in Excel.

Frequently asked questions

What does an ESG rating measure?

It measures a provider’s assessment of how a company handles environmental, social and governance issues, using that provider’s own methodology. MSCI describes its rating as industry-relative; Sustainalytics describes its rating as an absolute measure of unmanaged risk.

Is a higher ESG rating always better?

It depends on the provider. On MSCI’s AAA to CCC scale, AAA is the top rating. On Sustainalytics’ scale, lower risk is better, so negligible is the most favourable level.

Why do ESG ratings differ between providers?

Berg, Kölbel and Rigobon (2022) attribute about 56% of the divergence to measurement, about 38% to scope and about 6% to weights, across six raters with pairwise correlations of 0.38 to 0.71.

Can I average ratings from different providers?

Not without first putting them on a common scale and direction, as in the worked example, and considering whether relative and absolute measures should be combined at all.

Sources

  • MSCI, ESG Ratings: https://www.msci.com/data-and-analytics/sustainability-solutions/esg-ratings (read 2026-11-11).
  • Sustainalytics, ESG Risk Ratings: https://www.sustainalytics.com/esg-data (read 2026-11-11).
  • Berg, F., Kölbel, J. F., & Rigobon, R. (2022). Aggregate confusion: The divergence of ESG ratings. Review of Finance, 26(6), 1315–1344. https://academic.oup.com/rof/article/26/6/1315/6590532 (read 2026-11-11).

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · free to try

Ask about ESG Ratings: How They Are Built and Why Raters Disagree

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →