Tag Archives: Polls

Exploring Value Space

If you have enough of a following, Twitter polls are a great resource for exploring how people think. I’ve just finished asking a 8 polls each regarding 12 different questions that make people choose between the following 16 features, either in themself or in others:

attractiveness, confidence, empathy, excitement, general respect, grandchildren, happiness, improve world, income, intelligence, lifespan, pleasure, productive hrs/day, professional success, serenity, wit.

The questions were, in the order they were asked (links give more detail):

  1. UpSelf: Which feature of you would you most like to increase by 1%?
  2. Advice: For which feature do you most want a respected advisor’s advice?
  3. ToMind: Which feature of yourself came to your mind most recently?
  4. WorkedOn: Which feature did you most try to improve in the last year?
  5. UpOthers: Which feature of your associates would you most like to increase by 1%?
  6. City: To which city would you move, options labeled by the feature that people there are on average better on?
  7. KeepSelf: If all your features are to decline a lot, which feature would you save from declining?
  8. Aliens: What feature would you use to decide which civilization survives?
  9. Voucher: On which feature would you spend $10K to improve?
  10. World: Which feature of yours would you most like to improve to become world class?
  11. Obit: Which feature would you feel proudest to have mentioned in your obituary?
  12. KeepOthers: If all of your closest associates’ features will decline a lot, which feature would you save from declining?

Each poll gives four options, and for each poll I fit the response % to to a simple model where each feature has a positive priority, and each feature is chosen in proportion to its priority. The max priority feature is set to have priority 100. And here are the results:

This shows, for each question, the average number who responded to each poll, the RMS error of the model fit, in percentage points, and then the priorities of each feature for each question. Notice how much variation there is in priorities for different questions. Overall, intelligence is the clear top priority, while grandkids is near the bottom. What would Darwin say?

Here are correlations between these priorities, both for features and for questions:

Darker colors show higher correlations. Credit to Daniel Martin for making these diagrams, and to Anders Sandberg for the idea. We have ordered these by hand to try to put the stronger correlations closer to the diagonal.

Notice that both features and questions divide neatly into self-oriented and other-oriented versions. That seems to be the main way our values vary: we want different internal versus external features, and different features in ourselves versus others.

Added 20Jan: Some observations:

There are three packages of features, Impressive, Feelings, and Miscellaneous, plus two pretty disconnected features, intelligence and grandkids. It is striking that grandkids is so weak a priority, and negatively correlated with everything else; grandkids neither make us feel better, nor look impressive.

The Impressive package includes: attractiveness, professional success, income, confidence, and lifespan. The inclusion of lifespan in that package is surprising; do we mainly want to live longer to be impressive, not to enjoy the extra years? Also note that intelligence is only weakly connected with Impressive, and negatively with Feelings.

The Feelings package includes: serenity, pleasure, happiness, and excitement. These all make sense together. The Miscellaneous set is more weakly connected internally, and includes wit, respect, empathy, and improve world, which is the most weakly connected of the set. Empathy and respect are strongly connected, as are wit and excitement. Do we want to be respected because we can imagine how others feel about us, or are we empathetic because that is a “good look”?

There are two main packages of questions: Self and Other. The Other package is UpOthers, City, Aliens, and KeepOther, about what we want in associates. The Self package is Voucher, World, ToMind, WorkedOn, and Advice, about how we choose to improve ourself. UpSelf and KeepSelf are connected but less so, which I interpret as being more influenced by what we’d like others to think we care about.

KeepSelf and KeepOther are an intermediate package, influenced both by what we want in ourselves and what we’d like others to think we care about. Thus what we want in others is close to what we’d like others to think we want in ourselves. It seems that we are more successfully empathetic when we think about the losses of others, rather than their gains. We can more easily feel their pain than their joy.

Obit is more connected to the Other than the Self package, suggesting we more want our Obits to contain the sorts of things we want in others, rather than what we want in ourself. 

Note that while with features the Impressive and Feelings packages are positively correlated, for Questions the Self and Other questions are negatively correlated. Not sure why.

 

GD Star Rating
a WordPress rating system
Tagged as: ,

How To Pick A Quack: Data

How do we pick, or think we should pick, our experts? One clue comes from “How to pick an X” web guides. For 18 types of experts X, I searched for that phrase, and read the top 8 Google hits, noting all of the types of info clues mentioned in each guide. Here is the full table of results.

Here are the 25 most common clue types, sorted by the % of these guides in which each is mentioned:

Here are the 18 types of experts, sorted by the average number of clue types that their guides mention:

Looking at these tables, I hypothesized that guides might prefer to mention types of clues that we’d more want to use in making our choices, and that guides might mention more clues for kinds of experts where we worry more about choosing them well. So I’ve done a set of 16 Twitter polls to estimate these things for 16 types of experts and 16 type of clues.

Results to note:

  • Guides for 18 different types of experts vary by a factor of 3 in how many types of clues they mention.
  • The top 25 info clues vary by a factor of 12 in how often they are mentioned in guides.
  • While different clues are favored in guides for different types of experts, the overall pattern looks pretty random.
  • Only 7.8% of guides mention a top 25 clue directly sensitive to outcomes. (Ones marked in red above.)
  • The correlation between how many clues guides to X mention and how worried poll respondents are re pick X is strong: +0.40.
  • The correlation between how often guides mention a clue and how much poll respondents want to know it to pick is negative: -0.20. This is mainly because polls put the most weight on track records. My followers are probably less representative here, as that’s an issue I talk much about.

Guides do not often mention outcome-related clues, presumably as few customers attend to them. In general, we can’t tell if a type of expert X is a “quack”, where “better” versions don’t help customers much more with outcomes, by the kind of clues people use to pick X. Maybe most people can’t tell the difference.

So what explanations can you offer for any of the patterns you see?

Added: Here are the poll-based priorities each expert type and info clue: Continue reading "How To Pick A Quack: Data" »

GD Star Rating
a WordPress rating system
Tagged as: ,

Lognormal Priorities

In many polls on continuous variables over the last year, I’ve seen lognormal distributions typically fit poll responses well. And of course lognormals are also one of the most common distributions in nature. So let’s consider the possibility that, regarding problem areas like global warming, falling fertility, or nuclear war, distributions of priority estimate are lognormal.

Here are parameter values (M = median, A = (mean) average, S = sigma) for lognormal fits to polls on how many full-time equivalent workers should be working on each of the following six problems:

Note that priorities as set by medians are quite different from those set by averages.

Imagine that someone is asked to estimate their (median) priority of a topic area. If their estimate results from taking the product of many estimates regarding relevant factors, then not-fully-dependent noise across different factors will tend to produce a lognormal distribution regarding overall (median) estimates. If they were to then act on those estimates, such as for a poll or choosing to devote time or money, we should see a lognormal distribution of opinions and efforts. When variance (and sigma) is high, and effort is on average roughly proportional to perceived priority, then most effort should come from a quite small fraction of the population. And poll answers should look lognormal. We see both these things.

Now let’s make our theory a bit more complex. Assume that people see not only their own estimate, but sometimes also estimates of others. They then naturally put info weight on others’ estimates. This results in a distribution of (median) opinions with the same median, but a lower variance (and sigma). If they were fully rational and fully aware of each others’ opinions, this variance would fall to zero. But it doesn’t; people in general don’t listen to each other as much as they should if they cared only about accuracy. So the poll response variance we see is probably smaller than the variance in initial individual estimates, though we don’t know how much smaller.

What if the topic area in question has many subareas, and each person gives an estimate that applies to a random subarea of the total area? For example, when estimating the priority of depression, each person may draw conclusions by looking at the depressed people around them. In this case, the distribution of estimates reflects not only the variance of noisy clues, but also the real variance of priority within the overall area. Here fully rational people would come to agree on both a median and a variance, a variance reflecting the distribution of priority within this area. This true variance would be less than the variance in poll responses in a population that does not listen to each other as much as they should.

(The same applies to the variance within each person’s estimate distribution. Even if all info is aggregated, if this distribution has a remaining variance, that is “real” variance that should count, just as variance within an area should count. It is the variance resulting from failing to aggregate info that should not count.)

Now let’s consider what this all implies for action biases. If the variance in opinion expressed and acted on were due entirely to people randomly sampling from the actual variance within each area, then efforts toward each area would end up being in proportion to an info-aggregated best estimates of each area’s priority – a social optimum! But the more that variance in opinion and thus effort is also due to variance in individual noisy estimates, then the more that such variance will distort efforts. Efforts will go more as the average of each distribution, rather than its median. The priority areas with higher variance in individual noise will get too much effort, relative to areas with lower variance.

Of course there are other relevant factors that determine efforts, besides these priorities. Some priority areas have organizations that help to coordinate related efforts, thus reducing free riding problems. Some areas become fashionable, giving people extra social reasons to put in visible efforts. And other areas look weird or evil, discouraging visible efforts. Even so, we should worry that too much effort will go to areas with high variance in priority estimate noise. All else equal, you should avoid such areas. Unless estimate variance reflects mostly true variance within an area, prefer high medians over high averages.

Added 3p: I tried 7 more mundane issues, to see how they varied in variance. The following includes all 13, sorted by median.

GD Star Rating
a WordPress rating system
Tagged as: , ,

What Future Areas Matter Most?

I made a list of 44 possibly important future areas, and just did 22 Twitter polls (with N from 379 to 1178), each time asking this question re 4 areas:

Over next 30 years, changes in which are likely to matter most?

I fit the answers to a simple model wherein respondents either pick randomly (~26% of time) or pick in proportion to each area’s (non-negative) “strength”. Here are the estimated area strengths, relative to the strongest set to 100:

Some comments:

  1. The area with the largest modeling error is migration, so politics may be messing that up.
  2. Governance mechanisms looks surprisingly strong, especially relative to its media attention.
  3. The top 7 areas hold half the total strength, and there’s a big drop to #8. ~20% is in automation, AGI, and self-driving cars.
  4. 19 areas have strengths lying within about the same factor of two. So many things seem important.
  5. Relative to these strength ratings, it seems to me that media focus is only roughly correlated. Media seems disproportionately focused on areas involving more direct social conflict.
  6. Areas add roughly linearly. For example, biotech arguably includes life extension, meat, and materials, and pandemics, and its strength is near their strength sum.
GD Star Rating
a WordPress rating system
Tagged as: ,