FAQ

CSRHub overall ratings are based on four categories: Environment, Employees, Community, and Governance categories. Each category has three subcategories. Environment –Environmental Policy and Reporting, Energy and Climate Change, Resource Management; Employees -- Diversity and Labor Rights, Compensation and Benefits, Training Health and Safety; Community – Community Development and Philanthropy, Human Rights and Supply Chain, Product; Governance –Leadership Ethics, Board, Transparency and Reporting. Review Our data schema for more information.

CSRHub takes information from its data sources and transforms it into a 0 to 100 scale. We then weight the data sources, aggregate their scores, and then adjust them so that they fit together smoothly. See our homepage and the CSRHub Rating Rules for more details.

Fully rated entities have an overall rating. They also have all four category ratings and at least five subcategory scores. We only offer an overall rating when an entity has enough ratings sources (and those sources provide enough ratings information) to allow us to estimate a rating within one point at a 95% confidence interval. We currently offer overall ratings on about loading... entities.

We have received data from about loading... sources on more than 100,000 companies and organizations. Eventually, we hope our system will allow us to publish ratings on all of these companies—and many hundreds of thousands of publicly-held companies, privately-held companies, not for profits, and government organizations. We have enough sources and data to provide full ratings across the four categories of data we track (Community, Employees, Environment, and Governance) for over loading... companies. For the rest of the companies in our system, we often have only one source of data on a company or information on only a few areas of a company’s sustainability performance. In loading... cases, we believe this data is sufficient for us to publish partial ratings—data on one or more of our twelve subcategories. These partial ratings generally allow us to also provide at least one category rating for these partially rated companies. See the steps we take to determine which ratings to publish on our partial ratings page.

Not rated entities have data sources—so we know they are starting to generate expert opinions about their sustainability. However, we are not confident enough about where they stand within our ratings universe to offer a partial or full rating.

In some cases, we may indicate our best current guess of a subcategory or category rating. Please recognize that these are tentative scores only and that they are not included in the overall averages for the site. We have about loading...  entities in this category.

CSRHub takes information from its data sources and transforms it into a 0 to 100 scale. The higher the rating the better, with 100 as the best rating. These ratings are currently clustered around 50.

No. The average company rating is typically about 50. That is because in general, companies have a ways to go to become sustainable and socially responsible. You can read more about how we calculate ratings on our Ratings Methodology page.

Percentile rank shows where a company’s ratings lie among a broader group of companies.

Let’s say that there are five companies with ratings 20, 25, 50, 55, and 80. The company with rating 50 has a 50th percentile rank. The one with 20 has a 0 percentile and the one with 80 has 100%. The one with 25 gives a 25% rank and the one with 55 gets a 75% rank.

It is clear from this example that percentiles are not as accurate a measure of a company’s perceived performance as would be the rating itself. Further, there is a strong central tendency within our ratings. As a result, two companies that are separated by a single ratings point at the bottom end (for instance) of the ratings curve might be only a percentage point apart in their ratings. However, two companies with a one ratings point difference might be four or five percentage points apart if they are in the middle of the curve.

Only professional level subscribers can take advantage of this feature and see the history of a company's rating. Please see this page for a full description of how our date rollback system works.

The overall score is based on a weighted average of the category scores. The weights used come from each user’s profile settings. There is a non-linear translation of weight from the profile settings and the scale for each category weight is different. So, a “3.0”setting for Environment is not necessarily 2X as much weight as a “1.5”setting and the difference between 3.0 and 1.5 for Governance could be different from that for Environment. The scales we employ have been derived from study of the profile settings of our users. We know how many users set profiles at each level, so we know how rare (and therefore how heavily or lightly weighted) each setting should be.
You can read more about our profile research here: https://blog.csrhub.com/2012/01/moving-beyond-green-the-rainbow-within-csr.html.

Our goal is to provide consistent ratings of sustainability performance across a large number of companies. We do not want to be trapped into rating only public companies (e.g., by relying primarily on public filings). We also do not want to interpret and impose ratings directly ourselves (e.g., using our own team of analysts). We want our sources to be transparent (and they are). We expect our ratings to reflect the general perception of a company’s performance, rather than a company’s true performance.

We face several methodological issues, when we approach our rating task:

A. Each source has its own perspective and reporting schema. We currently have more than loading... different ratings elements in our system. Within the area of “Board performance,”we may have sources who report that a company’s board is “good,”“C+,”or “2.3;”that a board is “diverse;”that “X% of the board members are women;”that a board meets “Y times per year;”that “the board is (or is not) involved in the sustainability process;”etc. To make sense of this mess, we have to map each element we ingest into one of the 12 parts of our schema. To make the system “fair,”we try to pull in roughly equal numbers of elements and data items for each of the 12 parts.
B. No source covers all companies. We already cover loading... companies in loading... countries. Our goal is to cover hundreds of thousands of companies in every country in the world. We will never have a single standard set of data against which we could scale and adjust everything else.
C. Sources tend to be biased or to have discontinuous ratings distributions. Once we have mapped a source, we can compare the ratings it gives for a company against all of the other ratings we have for that company. We can quickly determine if a source is biased positively or negatively. Some sources have only one value (e.g., “Yes, the company does have a policy”). We adjust each source’s results to fit the overall distribution for each subcategory and then adjust the overall distribution to reflect the input of the sources.
D. Some sources are more accurate than others. We see this again through a comparison with our other sources. When a source has poor accuracy, we reduce its weight in our system.

The above processes involve massive amounts of computation—we are following a “Big Data” approach. It would be impossible to explain to an outside user exactly why ten sources gave a score of 46 for one company and two of those source plus six others gave a score of 53 for another one. Not only would the conversions appear arbitrary without the support of our analysis (we currently have loading... data points in our system), but the conversions carry relatively little information. The most valuable information is the data we input from the sources (which we share) and our inventory of which sources track each company (which we also share).

We feel our industry and country averages also contain value. Due to the nature of our approach, all of our ratings are accurate to within less than 1.8 points. (In other words, if one company has a score of 51 and another has one of 49, we are 95% confident that the two scores are different.)

Although we have more than loading... data sources and more than loading... pieces of data, we don’t have every piece of data on every company. We report the information we have. Where we are missing details, we leave the information blank. This can sometimes distort the rating for a company.

Our sources overlap with each other in thousands of places. This allows us to gauge the relative attitude of each source and adjust for their biases. As a result, when we have only a few sources available for a particular company, we can adjust those sources—based on all of our experience—and create ratings that are consistent and, we hope, fair. If the amount of data we have falls below a certain level, we do not publish a rating. There are approximately another loading... companies for whom we have partial information—but not enough to provide a rating. For more information on how we calculate ratings, please review the information on this page.

We do not ingest any data directly from companies. Instead, we use the ratings and opinions formed by our loading... sources as the input to our system. Our system’s goal is provide our users with the best possible estimate of how each company is viewed by its stakeholders. We do not need to collect every available fact about sustainability. Instead, we want to collect all of the opinions about a company’s performance and synthesize them into a broad, consistent, and comparable estimate of perceived performance.

There are many reasons we took this approach:

A. The available facts are not truly comparable between companies. For instance, one company may include supplier contributions in its carbon use. Another may include employee travel. A third could include the effect of its mix of power sources. A fourth might not include its subsidiaries. A normal user who sees these four sets of numbers cannot draw a conclusion from them about which company has better carbon efficiency. In contrast, our expert sources can draw conclusions based on this type of data. By combining the input of several expert sources (after removing any biases we detect in their methodologies) we get a clear signal about which company appears to have better carbon management policies.

B. Each stakeholder group has a different focus and interest. A governance source may approve of a CEO who carefully supervises all of her or his operations. An employee source may dislike the same CEO, because she/he is a micromanager who interferes in day to day operations. A labor source will care less about management style and more about whether or not the CEO allows unions to recruit members and participate in workforce management. Those who try to collect and report facts miss these nuances and obscure the fact that sustainability performance must be measured using a context of personal and moral values.

C. Some of our sources charge large amounts for the use of their data. They are willing to allow us to ingest it into our system, but would be out of business if we passed through all the detailed data they have labored to collect.

If a company improves its policies and puts more emphasis on sustainability, its score in our system should eventually rise. However, the speed with which this happens may depend upon how well the company communicates the changes it has made. It may also depend on how each of the various stakeholder groups views these changes.

Our tool does not tell you how a company is doing. It tells you how everyone else who shares your view of the world (as described in your profile) would think it was doing, if they had access to all loading... of our sources. Our tool is designed to help sustainability practitioners get feedback from the “opinion marketplace” on their company’s perceived performance.

No, since we do not attempt to collect data directly from the companies we cover (see above). If a company used the Global Reporting Initiative (GRI) framework for its CSR report, committed to the UN Global Compact, joined the Sustainable Packaging Coalition, became a member of the American Sustainable Business Council, etc., they might already be rated (at least partially) by CSRHub. As a company expands the scope of their sustainability work, they will be noticed by our sources and we will start to learn about your activities. This organic approach is the easiest way to join the sustainability metrics world.

You can use our system to discover competitors who rank well and then do your own analysis of how they have attracted attention and praise for their work. As an alternative, CSRHub could introduce you to one of the sustainability consulting firms with whom we have partnered. Our partners could review your internal systems and programs, benchmark your performance against that of your competitors, and then advise you of the best ways to disclose and promote your sustainability achievements.

Some issues don’t fall into our four category structure. For instance, whether or not a company tests its products on animals, does business in Burma, or is a good place for women to work may not be directly covered by ratings on a company’s governance, labor relations, etc. Special issues are things you feel especially strongly about. You can choose to make their presence in a company a positive thing (e.g., you could upgrade companies that support unions), a negative thing (e.g., you could downgrade the ratings of companies that are against the NRA), or drop the ratings entirely for those companies who are involved in an area you feel strongly about. If you choose “exclude” for a special issue, a company that has this issue will show no rating. It will also be taken out of the ratings for its geography and industry. This reflects the fact that some of our users feel so strongly about special issues that they are uncomfortable with seeing any rating for these companies. See a complete list of the special issues we track.

Percentile rank shows where a company’s ratings lie among a broader group of companies. Let’s say that
there are five companies with ratings 20, 25, 50, 55, and 80. The company with rating 50 has a 50th
percentile rank. The one with 20 has a 0 percentile and the one with 80 has 100%. The one with 25 gives
a 25% rank and the one with 55 gets a 75% rank.

It is clear from this example that percentiles are not as accurate a measure of a company’s perceived
performance as would be the rating itself. Further, there is a strong central tendency within our ratings. As
a result, two companies that are separated by a single ratings point at the bottom end (for instance) of the
ratings curve might be only a percentage point apart in their ratings. However, two companies with a one
ratings point difference might be four or five percentage points apart if they are in the middle of the curve.

Our data set goes back (monthly) to December 2008. Because we integrate data from other sources, our ratings lag and are backward-looking. However, our data set should offer a coincident signal—e.g., our data for January 1, 2020 should represent what the historic view was of the companies we covered, at that time.

None.

The data for May 2020 should be available in July 2020.

No.

We cover any entity (public-traded, privately-held, or non-for-profit/government) for which sustainability-related information is available.

We normalize at the indicator level. We map at that level on a many-to-many basis. We weight at that
level. We also weight at the source level. Overall, this gives us a lot of flexibility to accommodate the
variety of types of data we receive and the “shape” of their ratings curves. We do not “pick” sources. Our
system picks them for us via the weighting approach. Since each company has a different array of
available sources, the distribution of weight and importance of each source varies by company. We do
not direct companies and tell them to work on particular sources. But, we can tell them in general where
they are strong and weak and we can show them which sources within each area have more or less
influence on their scores. Through this, companies can create an action plan that we hope is primarily
based on changes in their policies and practices. They can then create a communication plan and share
their improved behavior with those sources who may be interested. Companies often use us to find new
sources—groups they feel they want to engage with. This is a non-linear walk with many branches and
options.

Our ratings can’t be normally distributed because they are bounded (do not go to infinity). We use instead a beta distribution model. Each subcategory has its own median value and the peak of each curve may be at that median, to the right of the median, or to the left of the median. Our system tends to keep each median near 50, but some are always above and some are always below. The shape of each subcategory rating curve is different.

Each subcategory’s evaluation is done separately. Each one is fit to its own beta distribution. We adjust
for consensus at the source level. For instance, if a source is a biased right (has all positive readings), we
will compare it against all other ratings in that subcategory. It could be that the source is only looking at
companies that are positively rated by everyone else. If so, we’d leave its ratings biased right. If a source
strongly correlates with all other sources, we up weight the strength of that source.