What if you could look up the correlation between just about any pair of human traits in seconds, as easily as you can do a Google search?
You'd be able to immediately find answers to questions such as:
How strongly are depression and anxiety linked?
Are agreeable people more altruistic?
Are extroverted people more charismatic?
Are smokers more willing to take risks?
What traits are linked to being a psychopath?
Do women tend to be more spiritual than men?
Are religious people more likely to value having children?
Until now, this was not possible. We're very excited to tell you that we've just launched a project that allows you to answer questions like those and close to a million others! It's called PersonalityMap, and you can use it now, for free! You can also read on to learn more about how it works and to get the answers to the questions about the relationships between human traits that we've listed above.
What PersonalityMap Does
This free tool revolves around two things: first, we have compiled a wide variety of real psychological datasets and survey results, totaling over 1 million correlations, which we're making available through PersonalityMap. As far as we know, this is the biggest database of correlations that has ever existed - and it’s free!
Second, we trained a sophisticated machine-learning model on that data. This way, PersonalityMap not only provides users with the actual correlations found in existing studies but is also able to predict entirely new correlations through said machine-learning model. These predictions are not real study findings - instead, they are derived from the (sometimes) incredible power of machine learning to draw connections and make inferences (in some cases even ones that human minds cannot - after all, no person has poured through a million correlations like our model has). They are machine judgments about what human traits are likely to correlate (and by how much). Of course, you may wonder, why is all this useful?
The Digital Petri Dish
In the past, if you wanted to know how two human traits related to each other, you’d have to hope that a study on the topic had already been conducted. Even then, unless a plainly worded article had also been written about it, you'd have to expend considerable effort to find that study and extract the relevant information from somewhere deep in the paper. And the paper would probably have been behind a paywall. That’s if you were lucky. If you were unlucky, you’d find that no study existed yet that looked at the exact relation you’re interested in – in which case, the only way to get the answer would be to run your own study. And although we've been involved in making running studies easier and faster, with sister projects like GuidedTrack for building studies and Positly for recruiting participants for research, we know that, realistically, few people are going to run their own study.
So, what can you do?
It turns out, there's an interesting way in which psychology and social sciences are at a disadvantage to biology: while there are many biological studies with human subjects, biologists have the much faster option of what’s called ‘in-vitro’ experimentation. Instead of studying actual humans, which is not only costly but also comes with all kinds of other limitations and constraints, they can just conduct experiments in a Petri dish! This may yield less reliable results, as it’s further removed from the thing you’re usually ultimately interested in – effects on actual humans – but it’s nonetheless an invaluable tool that has vastly accelerated biological research. With in-vitro experimentation, scientists can run many more experiments in less time, which is a good first step in generating and testing initial hypotheses of things to test on actual humans afterward.
Well, now PersonalityMap provides a digital Petri dish for human psychology. It gives you access to over a million human correlations, spanning personality, demographics, behaviors, and beliefs!
Testing PersonalityMap
To assess our machine-learning model’s predictive performance, we teamed up with psychologists and tested its ability to predict personality correlations on pairs of personality statements that the model had not been trained on. We released the results in a new study. Our model outperformed 100% of the 254 non-experts and more than 99% of the 272 psychology experts (for questions that the model had never seen before).
Now, let’s look back at the examples from the beginning of this article. Here are the real correlations (from real studies) retrieved from PersonalityMap:
Question | Answer from PersonalityMap | PersonalityMap Study Correlations |
How strongly are depression and anxiety linked? | Extremely! | |
Are agreeable people more altruistic? | Yes! | |
Are extroverted people more charismatic? | Yes, strongly! | |
Are smokers more willing to take risks? | Only barely/negligibly! | |
What traits are linked to being a psychopath? | Based on predictions from our machine-learning algorithm: Being disagreeable, unempathetic, unkind, unemotional, selfish, cold, money-focused, vengeful (source) | |
Do women tend to be more spiritual than men? | A little bit! | |
Are religious people more likely to value having children? | Yes! A moderate amount. |
It's important to remember that these really are just correlations. Correlations are useful for prediction and understanding. However, it’s often tempting to jump to conclusions about causation (i.e., assuming that if A and B are correlated, then A causes B), but as we’ve described before, this is often misguided. Many correlations can be explained by confounding variables (e.g., C causes both A and B separately), reverse causation (B causes A), or other factors (like cyclic causation or sampling error).
How Does PersonalityMap Work?
PersonalityMap is powered by an advanced, novel machine-learning algorithm of our own design. We’ve compiled a huge variety of datasets from psychological studies and used these to train the machine-learning model to predict correlations between human traits, including personality traits, beliefs, demographics, behaviors, and more.
When using the machine-learning model to make predictions, you generally get the most reliable results when looking at the relations between statements that have been used before in studies – these are automatically suggested while you type into the website’s search bar:
However, in principle, you can even go beyond that and type in arbitrary questions or statements (though the less similar those statements are to those in the included studies, the less accurate the predictions about it will be). The model will (very loosely speaking) compare whatever statement or question you type to similar statements that it knows correlations for (from studies) and make its predictions based on that. Because it has seen over 1 million real correlations during training, it is often able to predict novel correlations well whenever they are close enough to what it knows.
So you can even gain some (more speculative but still often informative) insight into the correlation of two things that have never (even once in history) been part of the same study – or any study at all, for that matter! For instance, “I’m a narcissist” is a statement that has not directly been part of any studies in the training set. Yet, the model can predict a wide variety of correlations (such as agreeing with the statement "I make myself the center of attention").
What Are Some Use Cases of PersonalityMap?
PersonalityMap allows you to get answers to (and predictions about) a variety of questions about human psychology, whether you're a curious layperson or a seasoned researcher. But now, let's dive into some example use cases, starting with some that may be interesting to anyone and then gradually getting deeper into research territory.
For the use cases below, we're using the full spectrum of PersonalityMap's features. This also means that many of these use cases utilize AI predictions to varying degrees (e.g. when looking at the "Factors" or "Linear Regression" tabs, or whenever a predicted correlation has no corresponding study correlation). However, even if you're very skeptical of AI tools and would rather avoid relying on AI predictions altogether, you can still benefit from most of these use cases: just focus on the "Most Correlated" entries that show a study correlation on the right, or search for particular study questions in "Predict Correlation" that come with a study correlation. Even without relying on AI, PersonalityMap empowers you to access its huge (and growing) database of correlations in a matter of only a few clicks!
Case 1: Understanding a trait (such as psychopathy or depression)
Perhaps you are interested in understanding some personality trait or psychological condition. To pick up an example from the beginning of this article, let’s say you are interested in the most common traits of psychopaths. By searching "psychopath," you can identify that the database contains the statement, “I’m a psychopath, not an empath.” The “Most Correlated” tab and the “Factors” tab provide some insights into what this statement tends to correlate with most strongly and what factors it is predicted to correlate with. In this case, this happens to include traits such as high disagreeableness, low compassion, and being male. It also correlates with statements such as “I’m cold, not warm,” “I’m logical, not emotional,” and “I’m vengeful, not forgiving”
Or maybe you have a friend who’s depressed, and you don’t know much about depression but would like to understand their condition better. With PersonalityMap, you can quickly find out that depression correlates strongly with, e.g., feeling hopeless, negative self-talk, worrying a lot, and having very little energy. This might help you think of more informed strategies for supporting your friend.
Case 2: Sanity-checking claims or beliefs (such as that female study participants are more revenge-seeking than male ones)
You may hear others confidently make claims that don’t seem necessarily true to you. Maybe a friend of yours claims that women generally tend to be more manipulative and revenge-seeking than men. Questioning this, you might then find that PersonalityMap contains items like “manipulativeness” and “I do things out of revenge”, and check the Factors tab. And there, you would see that your friend’s claims don’t hold up: being female is (weakly) negatively correlated with both manipulativeness and revenge-seekingness.
Case 3: Looking up a relationship between traits (like anxiety and irritability)
Let’s say you’re formally researching (or merely have a bet with a friend about) the relation between irritability and anxiety. Are anxious people more irritable, on average? One way to check this is to pick one of the existing questions in our database relating to irritability, e.g., “I tend to be irritable,” and then look for correlations with anxiety-related statements. One big correlation you would find is with the statement, “I see myself as anxious and easily upset”:
The predicted correlations (from the machine-learning model) are shown along the left (r=0.66 in this case), whereas the correlations from real studies are on the right (r=0.64). You can also see that the machine-learning model is still able to make predictions in cases where there is no actual correlation from a study, as in the case of the third row.
Note: the values of the predicted correlations in PersonalityMap may change slightly over time as we keep improving the model, thereby subtly improving its predictions. Similarly, you might find more study correlations over time, whenever we add new studies to our database.
Case 4: Generating Hypotheses (such as predictors of criminality)
PersonalityMap makes it easy not only to verify concrete hypotheses you already have but also to freely explore all the correlations for a trait or question that you may not have ever considered. This makes it easy to come up with new hypotheses.
Maybe you’re interested in predictors for criminal behavior. In this case, you could start with “How many months in total have you spent in jail or prison?” and explore its correlations. Among these, you would find some interesting ones, such as:
“I like attracting attention” (study correlation: 0.26)
“I enjoy watching violent sports” (predicted correlation: 0.17)
Knowing of these correlations and keeping in mind that correlation doesn't necessarily imply causation (does watching violent sports create violence, do violent people like watching violent sports, or is there some other explanation for the correlation?) you could then come up with more concrete hypotheses of how and why these are correlated, which you could then test in a study to confirm that they hold up.
Case 5: Generating causal hypotheses and ruling out some causal pathways
We mentioned earlier that PersonalityMap is all about correlations, and thus, one needs to be very careful when making any inferences about causation. However, there are still some ways in which it can provide some evidence for and against causal claims.
First, as we discussed in a different piece, causation without correlation can exist but is quite rare. This means that when you find a correlation between two factors that is close to 0, this is evidence (but not proof) against a meaningfully strong causal link between the two. It's often the case that correlated things are not causally linked – but it's much less often the case that uncorrelated things are actually causally linked. So, a lack of correlation is evidence of a lack of causation!
Second, suppose you have a concrete causal hypothesis that two factors, A and C, are causally linked. However, your theory is that they are not linked directly but through some intermediate factor B. For example, extraversion is linked to increased happiness: agreeing with the statement “I am happy with my life” has a correlation (from a real study) of 0.24 with the trait of extraversion. But your theory is that this happens exclusively through spending more time with friends (i.e., extroverts spend more time with friends, and that's why they are happier).
To check this, you would first search for a suitable question related to the end of your causal chain, in this case, happiness. Going with the statement “I am happy with my life” to represent happiness, you would then move on to the “Linear Regression” tab, where you can look for suitable questions for the other two factors. There are a number of extraversion-related statements in the database, one being “extraversion (Big 5 personality composite score)”. After adding this, we obtain a predicted linear regression coefficient of 0.21, providing evidence for a link between extraversion and happiness.
Next, we look for a statement suitable for reflecting time spent with friends. A suitable statement might be “I am satisfied with the quality and quantity of my social interactions and friendships.” Adding it to the linear regression, we would expect that, if our theory is right, the predicted regression coefficient of extraversion drops from its original 0.21 to a value much closer to 0 – because then the friend factor alone would be sufficient to predict the increased happiness, with extraversion adding little extra information beyond having many or good friendships. Indeed, the coefficient is more than cut in half, moving from 0.21 to 0.09. This provides some evidence that your theory may be partially correct: part of the increased happiness of extroverts (albeit not the whole effect) may be due to better or more friendships!
Case 6: Deciding how many participants to use in a study (i.e., better-informed power calculations)
A common problem of studies is low statistical power – that is, the study doesn't have enough participants in it to reliably detect the size of effects they are likely to find. This can happen if either no power analysis (a statistical approach for deciding how many participants the study should involve) is performed while planning a study or when researchers expect their study to find an optimistically large effect. This is problematic because even when real effects exist, an underpowered study may not find it and, therefore, falsely conclude that no effect exists (i.e., such studies are prone to "false negatives").
PersonalityMap does not predict effect sizes directly, but for many studies that are investigating questions that PersonalityMap can make predictions for, you can convert its correlation estimates into effect size estimates and, therefore, use them to estimate the number of participants to include in your research.
Our Vision For PersonalityMap
PersonalityMap is one in a series of sister projects we’ve launched, with the ambition of improving social sciences and psychology, and making them more accessible. Many of these are free to use and all are accessible in your browser, with no need to download anything. The full list now includes:
GuidedTrack streamlines the creation of advanced surveys and studies end to end, including complex branching logic and randomization, automated participant emails, and integration of AI into studies. (This is what Clearer Thinking uses to build its tools.)
Positly has the goal of making it easier, cheaper, and faster to recruit study participants from all over the world.
We launched Transparent Replications in an effort to help improve the replication crisis in psychology by running rapid but high-quality replications of new psychology papers published in top journals.
Now, we're launching PersonalityMap to help accelerate the discovery and verification of important insights about humanity.
Our primary goal with PersonalityMap is to enable researchers and anyone with an interest in human psychology to rapidly look up known answers to psychological questions, generate new hypotheses, make predictions about important topics, and check their ideas (before running later confirmatory studies).
Over time, we plan to add more and more studies to our system, leading to a wider range of questions that can be explored with Personality Map, from health to beliefs to behaviors and beyond. We expect that an increasingly diverse set of datasets will also continue to make our machine-learning model more and more accurate when making predictions on questions for which no studies have ever been conducted.
We'd really love for you to try PersonalityMap out for yourself (it's free) and share it with others who you think may be interested. We'd also be really interested in any feedback you have if you try it out – just click in the upper right-hand corner of the site to give feedback.
Here's a button to give PersonalityMap a try: