I took this test for the Big Five twice at an interval of two years. The Big Five traits have been discussed on LW, mostly in a favorable light.
My scores were as follows, showing quite significant variations:
O65-C52-E42-A44-N37 on 12/06/2010
O30-C41-E31-A44-N32 on 29/06/2012
As a result I have updated quite a bit away from lending much credence to this particular self-administered Big Five test; either the test itself is flawed, or self-administration in general doesn’t ensure stability of the measurements, or the Big Five theory has a problem.
I agree with gwern that there’s not really much variation apart from the openness domain. It’s a bit dangerous to use percentile rankings on internet assessments for longitudinal studies, though – you never know if the norms have been changed, or if you have been normed against a different population when retaking the test due to differences in, e.g., IP address or age. It would be best to record the raw scores as well, if possible.
The test you linked to was created by Gosling et al. (2004) for a study on web-based Big Five tests. (I found it funny that this test was also created for the same study – ignoring the substantial differences in… decoration, they should give similar results.) The inventory in that test is the Big Five Inventory (BFI) (most recent reference: John et al., 2008); it’s quite widely used.
I recommend the IPIP-NEO for anyone who wishes to do a self-assessment for the Big Five. That link provides two versions: Goldberg (1999) developed the original 300-item inventory, and Johnson (2011) shortened that to a 120-item inventory. For those who have time, the 300-item version is psychometrically superior, as expected. There are two main advantages to the IPIP-NEO, compared to alternatives like the BFI: (1) It was designed to correlate with the commercially-distributed NEO-PI-R, which remains the most popular inventory in the literature, (2) It gives percentile rankings on the 30 facet-level scales as well as the 5 domain-level scales in the NEO-PI-R (an example report).
(I recently spent ~2 weeks doing a literature review of personality psychology, with a brief focus on internet self-assessments for the Big Five. For a very brief overview of the field, I recommend Robins and Donnellan’s (2009) article in The Corsini Encyclopedia of Psychology. For an up-to-date review of the Big Five, I recommend McCrae’s (2009) article in The Cambridge Handbook of Personality Psychology.)
This test seems to be measuring your self-image, not your behavior.
Consider the difference between asking, “Do you see yourself as someone who starts quarrels with others?” and following the person around for a week and seeing if they start quarrels with others.
I’d like to know how well-calibrated people’s self-images are. It seems to me that for some variables, many people’s self-images are very poorly calibrated.
Scud: I never said it had to be wrong. I’m just saying that if you accept that there are no Big 5 SNPs, then the explanation must be either:
1) Personality is independent of genetics
2) The Big 5 is a poor measure of personality
or 3) We just haven’t found the personality SNPs yet because our sample sizes are too small or our stats aren’t clever enough.
I believe that the original post mistakenly implies that height SNPs have been replicated. Similarly, IQ SNPs haven’t been replicated. Height and IQ have much higher retest correlation and much higher heritability than any big 5 factor. So option 3 is pretty reasonable. Option 2 might be true, but it doesn’t help explain the data.
Fitness relevant variation should be hard to detect, because selection is purifying it. If personality is fitness-neutral, then it might be easier to detect.
Take height. Though health and nutrition can affect stature, height is highly heritable: no one thinks that Kareem Abdul-Jabbar just ate more Wheaties growing up than Danny DeVito. Height should therefore be a target-rich area in the search for genes, and in 2007 a genomewide scan of nearly 16,000 people turned up a dozen of them. But these genes collectively accounted for just 2 percent of the variation in height, and a person who had most of the genes was barely an inch taller, on average, than a person who had few of them. If that’s the best we can do for height, which can be assessed with a tape measure, what can we expect for more elusive traits like intelligence or personality?
The genome is the ultimate spaghetti code. I would not be at all surprised if some genes code for the direct opposite characteristics in the presence of other genes. It is going to take more than just running relatively simple correlation studies to untangle the functions of most genes. We are going to have to understand what proteins they code for and how they work.
I took this test for the Big Five twice at an interval of two years. The Big Five traits have been discussed on LW, mostly in a favorable light.
My scores were as follows, showing quite significant variations:
O65-C52-E42-A44-N37 on 12/06/2010
O30-C41-E31-A44-N32 on 29/06/2012
As a result I have updated quite a bit away from lending much credence to this particular self-administered Big Five test; either the test itself is flawed, or self-administration in general doesn’t ensure stability of the measurements, or the Big Five theory has a problem.
I agree with gwern that there’s not really much variation apart from the openness domain. It’s a bit dangerous to use percentile rankings on internet assessments for longitudinal studies, though – you never know if the norms have been changed, or if you have been normed against a different population when retaking the test due to differences in, e.g., IP address or age. It would be best to record the raw scores as well, if possible.
The test you linked to was created by Gosling et al. (2004) for a study on web-based Big Five tests. (I found it funny that this test was also created for the same study – ignoring the substantial differences in… decoration, they should give similar results.) The inventory in that test is the Big Five Inventory (BFI) (most recent reference: John et al., 2008); it’s quite widely used.
I recommend the IPIP-NEO for anyone who wishes to do a self-assessment for the Big Five. That link provides two versions: Goldberg (1999) developed the original 300-item inventory, and Johnson (2011) shortened that to a 120-item inventory. For those who have time, the 300-item version is psychometrically superior, as expected. There are two main advantages to the IPIP-NEO, compared to alternatives like the BFI: (1) It was designed to correlate with the commercially-distributed NEO-PI-R, which remains the most popular inventory in the literature, (2) It gives percentile rankings on the 30 facet-level scales as well as the 5 domain-level scales in the NEO-PI-R (an example report).
(I recently spent ~2 weeks doing a literature review of personality psychology, with a brief focus on internet self-assessments for the Big Five. For a very brief overview of the field, I recommend Robins and Donnellan’s (2009) article in The Corsini Encyclopedia of Psychology. For an up-to-date review of the Big Five, I recommend McCrae’s (2009) article in The Cambridge Handbook of Personality Psychology.)
This test seems to be measuring your self-image, not your behavior.
Consider the difference between asking, “Do you see yourself as someone who starts quarrels with others?” and following the person around for a week and seeing if they start quarrels with others.
I’d like to know how well-calibrated people’s self-images are. It seems to me that for some variables, many people’s self-images are very poorly calibrated.
That doesn’t seem to be much of a variation, except for your Openness—and without knowing more about you, I couldn’t say whether that’s good or bad...
http://neuroskeptic.blogspot.com/2012/03/personality-without-genes.html
The author in the comments:
Also interesting: http://blogs.discovermagazine.com/gnxp/2012/06/heritability-of-behavioral-traits
I believe that the original post mistakenly implies that height SNPs have been replicated. Similarly, IQ SNPs haven’t been replicated. Height and IQ have much higher retest correlation and much higher heritability than any big 5 factor. So option 3 is pretty reasonable. Option 2 might be true, but it doesn’t help explain the data.
Fitness relevant variation should be hard to detect, because selection is purifying it. If personality is fitness-neutral, then it might be easier to detect.
I forgot about that:
From Steven Pinker in http://www.nytimes.com/2009/01/11/magazine/11Genome-t.html
The genome is the ultimate spaghetti code. I would not be at all surprised if some genes code for the direct opposite characteristics in the presence of other genes. It is going to take more than just running relatively simple correlation studies to untangle the functions of most genes. We are going to have to understand what proteins they code for and how they work.
Yes, that’s all people claim, and I believe even this 2% claim failed to replicate.