Formalizing Transparency

Introduction / Why I Wrote This

The charity sector has made a big deal out of ‘transparency’ recently. Here at 80,000 Hours (I work there now) there’s a lot of talk about how we and similar organizations could become more ‘transparent’. One discussed aim is to be more ‘transparent’ than Givewell, which in general is the gold standard of ‘transparency’ within the small field.

I’ve had discussions with many people who had many ideas on this topic. Eventually it dawned on me that I had little idea of what ‘transparency’ even meant. On further reflection, I don’t think many of these people did either. Some organizations have quantized transparency indexes into rubrics (see that for the World Bank, for example), but it wasn’t obvious what the rubrics were trying to estimate. I desired an ideal quantification of transparency that can be approximated in rubrics, rather than rubrics that match a general intuitive understanding of transparency.

I should also be clear that I am extremely demanding in what I want in definitions. I’m an engineer by training, I want to work with things I can quantify and model in the same ways I do Force and Velocity. Important ideas considered currently ‘soft’ or non-technical feel like challenges to define better. I realize that actual measurements may be extremely difficult, but this doesn’t mean that we can’t make estimates of substantial value. This definition attempt is me attempting to take my intuitive understanding of this word ‘transparency’ and match it to something technical and clear.

I don’t claim to have solved this problem, but I believe I’ve taken a few interesting steps. Hopefully we will eventually create a transparency rubric for effective altruist organizations, and I now have a much better understanding to begin with.


Quite simply, I am currently using the working model that transparency is defined as, > The total expected value of information delivered by a system, about itself.

I believe this definition is intuitive and almost quantifiable. The use of this definition should lead to more clarity regarding the term.

While I believe that this definition almost captures the ‘essence’ of what we mean as ‘transparency’, I also believe that a thorough evaluation of this essence leads to several facts that may be quite unintuitive.

#1: Information quality and accessibility are more important than quantity.

A few thought experiments

1. Reverend Robert Shields had a diary of 37.5 million words. The following is an example: > > * April 18, 1994
> * 6:30-6:35: I put in the oven two Stouffer’s macaroni and cheese at 350°. > * 6:35-6:50: I was at the keyboard of the IBM Wheelwriter making entries for the diary. > * 6.50-7.30: I ate the Stouffer’s macaroni and cheese and Cornelia ate the other one. Grace decided she didn’t want one. > * 7.30-7.35: We changed the light over the back stoop since the bulb had burnt out.

Every detail of Roberts’ thoughts seems to be burried within this text. Does this make him completely transparent? Would this change depending on the format (physical book, online pdf, online spreadsheet)?

Compare that to Stephen King who published On Writing, a rather short (320 page) autobiography and set of lessons he learned through that time. Would Stephen King have been more transparent had he instead wrote down his life information minute by minute?

2. Say there were a ‘perfect’ and ‘perfectly trusted’ being who evaluated nonprofits. They would rate them a ‘1’ if the nonprofit was the “best” nonprofit, and a ‘0’ otherwise. Rather than having any mention of what they do, many nonprofits decide to only display this number on their website (it would be a very simple website, perhaps with a url chosen to obfuscate their cause, like charity34234235.com). Are these charities being transparent to those trying to find the “best” charity? Could they possibly be more so?

My Take

‘Transparency’ is only loosely associated with the amount of information available. It doesn’t matter how much information is available if it’s useless, and extra information could easily lead to obfuscation of what matters.

When asking for a important organization details about different charities, I hear a lot of “Oh, we mentioned in the middle of a blog post a few months ago.” The conversation typically ends there. But on reflection, this information being difficult to find may be much more important than the fact that it exists somewhere. The vast majority of people who may be interested in it will never access it. I’m fortunate to be in a place where I can physically speak to many of the knowledge experts. Most people aren’t.

#2: Transparency is typically relative to a specific purpose.

If a charity had complete financial records of absolutely all expenses and revenues, would it make it completely transparent? What if it had this material presented cleanly with pretty graphs and signatures from several top accountants who have thoroughly reviewed it and are willing to cite it as clean and it’s filed taxes as completely legal?

This is a bad question because “transparency” isn’t well defined. This organization may have amazing financial transparency, or transparency for the purpose of understanding if it is partaking in sketchy or illegal accounting practices. But this is clearly a different kind of accountability than most potential donors would hope for. Many registered charities are required by law to have their main finances accessible on their websites, but this does not make them incredibly transparent. Most people would have little idea or interest to read such documents.

It’s obvious that there are different “types” or transparency, but I’ll clarify that to say there are different “purposes” for transparency to provide. There’s the “purpose” of transparency to understand if a nonprofit is committing financial fraud. There’s the “purpose” of transparency to understand if a nonprofit is effective vs. other nonprofits when it comes to cause-neutral returns, or local town benefits, or social status improvements.

There’s just no such thing as an organization being objectively “transparent”. “Transparency” only exists in relation to a specific purpose.

Going further, it seems like what we think of as “transparency” can be narrowed down to something similar to the value of information for specific decisions.

Separating ‘Global’ Transparency from ‘Local’ Transparency

While transparency typically refers to specific purposes, this doesn’t mean that the total amount of transparency that an organization is unbounded with an infinite number of possible purposes. Rather I assume that there are probably a few purposes that could bring the substantial majority of benefit. For instance, a 1-year old child’s communications would provide informational of tremendous value to their own health and their parent’s wellbeing. These could also be used to provide interesting data in larger studies, but the value from one individual baby on these may be much less in comparison. It could also provide moderate interest to a person considering later having a child, but this would be considerably less value potential still.

So a ‘total transparency’ score may be able to be estimated based on the sums of the values from each type of transparency. This adds complexity because it would seem to demand information about the decision maker categories in relation to each other. Of course, these can be simplified a lot by just choosing the top three categories and having a basic ranking among them.

This definition is not too different than the typical intuitive one. A banking organization would be seen as transparent primarily if it provides useful information that the government is looking for, and this information value is likely far more than that of most other domains.

#3: Transparency includes marketing and branding.

I’ve heard a lot of business and nonprofit pitches. They are almost universally completely positive about themselves.

The “business pitch” is exactly the antithesis of a “transparency report” or “self-evaluation” of any kind. Clearly, prospective donors would get more “value of information” from a completely unbiased and honest opinion than a sales pitch. But should we judge these pitches when analyzing an organization’s ‘transparency’? Or should “marketing materials” be held exempt from any ‘transparency’ analyses?

My uncomfortable answer is no. Good business pitches are bad transparency. It’s a sacrifice that we seem to be willing to endure, perhaps that we should. If this means that essentially no organization can be completely transparent, then so be it.

#4: Negative transparency is possible and common.

If transparency is related to the value of information provided, then 0 transparency would mean 0 information. There is definitely information that is worse than 0 information, especially to regular people (as opposed to rational super-beings).

A simple example is smoking. Someone goes to a website to learn information about smoking, and the website claims authority while selectively providing evidence for smoking. The information to the reader results in a higher likelihood of making a worse decision, so this information would have substantially negative value.

Likewise, if a nonprofit could use a terrible argument to convince an otherwise unbiased individual to donate to a worse cause then they would have otherwise (“wow, apparently visits to local prisons has been shown to scare select children from committing crimes!”), then that information has provided them with negative value.

#5: Internal transparency is often more important than external transparency

#6: Transparency should be judged as a comparison to optimality rather than as an absolute value.

So far I’ve stated that transparency relates to the value of information provided by an system. Yet, large organizations can easily provide extra informational value through their scale as opposed to anything we would associate with transparent efforts. It seems intuitive that small groups should have a “fair playing ground” in comparison.

I think that as opposed to thinking of transparency as directly relating to the absolute value of information provided, it makes more sense to think of it in an ‘percentage of optimal’ sense than a ‘total benefit’ one. This way when we say an organization is doing a ‘good job’ being ‘transparent’, it will always mean that they’re doing relatively well given what they are capable of.

A percentage of optimality seems a bit depressing as I’m sure that ‘optimal information value’ would be orders of magnitude higher than what most, if any, organizations currently provide. So perhaps rather than being compared to optimality, organizations can and should be compared to a bar significantly lower and more realistic. Perhaps different standards should created with top bars at different amounts of optimality. For instance, most existing transparency rubrics that I’m aware of probably are quite far from ‘optimal’, but still are quite useful and likely would correlate decently well with my working definition of transparency.

It should also be noted that because of diminishing marginal returns, reaching completely ‘optimal’ transparency, or anything like it, would likely reduce the efficiency of an organization towards its main goal. Or, an organization seeking anything other than to maximize transparency will not and should not get near optimizing transparency. Rather they should see that initially high benefits of transparency are worth some reasoned level of effort and achievement.