A while ago I had intended to write a post about my identity and all that I felt I had learned about who I am over these last few years but amongst other things it got sidelined. Then this week I was watching the second episode of the BBC documentary series Secrets of Silicon Valley in which the use of Facebook and social media for psychometric testing was discussed and it has prompted me to finally finish this post.
Reasons For Analysing And Sharing My Personality
My primary aim when I started writing was improving the accuracy of other people’s perception of my identity i.e. I wanted to ensure that the judgements others were forming of me were accurate representations of who I actually am.
This was something of particular importance whilst I was in the process of applying for jobs. Larger recruitment services often already use personality profiling and psychometric testing to try and identify suitability of a candidate for a role anyway, and smaller companies that lack the resources to do this through technology processing large amounts of the data still need a method to judge suitability, but just process using their own brains to assess who a candidate is through processes like interviews and reading their LinkedIn profile.
I have no control over what processes are used, by human or machine, although I can influence the data available to input in. However, there is an almost infamous lack of feedback in the job application process i.e. information about why I did or did not get the job, so I am unlikely to know what those conclusions reached were in order to judge for myself how accurately they represent me so that I can improve the data available to improve the accuracy of the conclusions reached. This is especially important given that I am statistically very likely to be subject to a lot of subconscious biases that give me a much higher chance of being misjudged.
Therefore I thought perhaps if I could find results of these processes myself that I believed to be very accurate then I could simply communicate that information directly.
Methods Of Analysing And Defining My Personality
Personality profiles are not the only defining features of identity, but to improve the economy of the psychometric processes used by corporations for things like marketing and recruitment there is a lot of work on trying to identify the minimum required variables needed to define an identity, and a lot of that centres around linking back other features to personality profiles both in terms of correlation for statistical analysis, and causation for greater understanding and control.
I am not qualified in the relevant fields of study involved in researching identity but I did not think it necessary to look through all the possible features and their relationships in too much depth, focussing instead just on those I believed might be used in or influencing technology processing my online data to try and define me in situations like job profiling assessments.
‘The Big Five’ / OCEAN Model
This is currently the most common method of defining personality, in terms of five variables (which can each be broken down into sub-variables for more detail too):
I completed a “Revised NEO Personality Inventory” test as part of a psychology experiment I was paid to participate in in December 2015, although I was not aware (as part of the experiment) at the time what questions I was answering belonged to this and which were assessing other things (they were screening people for selection for further experiments but I still do not know what for). I was provided with very detailed feedback from this.
Given the complexity and consequently the large amount of data needed to build this personality profile though, most companies cannot afford to ask people to complete a very long time consuming questionnaire to get sufficiently reliable results. However, it is believed that these values can be calculated from statistical analysis of the large amounts of information that are already available to these companies through the likes of your social media.
So Wikipedia describes the Myers Briggs personality model:
The MBTI was constructed by Katharine Cook Briggs and her daughter Isabel Briggs Myers. It is based on the typological theory proposed by Carl Jung,who had speculated that there are four principal psychological functions by which humans experience the world – sensation, intuition, feeling, and thinking – and that one of these four functions is dominant for a person most of the time.
So the factors measured here are:
- Extraversion vs Intoversion
- Sensing vs Intuition
- Thinking vs Feeling
- Judging vs Perceiving
Often the results are defined in terms of an acronym of the letters of the names of dominant end of each spectrum. Generally Myers-Briggs is not considered as consistent or relevant for scientific research as the OCEAN model, but despite that it is still very commonly used, especially for marketing and recruitment, so I gave it a look over.
Social Media Psychometrics And Quizzes
Apply Magic Sauce (AMS) (https://applymagicsauce.com/) is an API, with a demo on their website, developed by the University of Cambridge to apply algorithms to social media to derive an OCEAN personality profile and other identity features. You can try out the demo of it to analyse your Facebook and Twitter profiles for free and so I headed over to the site to see how one of the biggest algorithms defines my identity based upon my online profiles.
I tried the demo on their site with just my Facebook profile alone initially and then repeated the analysis with just the data from my Twitter profile alone, and then again with the data from both platforms.
I then discovered was that the AMS website provided an additional option to complete a self reporting study online to input instead and so I completed this too.
Then I found on another University of Cambridge website (https://discovermyprofile.com) several quizzes that used and tested the relationships between personality other identity features such as generating an OCEAN model profile based upon opinions on music or brands. I felt these were important to look at because they could potentially give an indication of how different specific variables influenced or potentially form part of the more generalist algorithms that companies are using and so giving my a greater understanding of how to influence them.
The Results Of My Identity Analysis
I have put the detailed results of each feature on a separate page if anyone is interested in looking over them, and then made infographics of the key results below:
The Big Five / OCEAN Model
From Facebook alone I discovered firstly that I do not have sufficient “likes” to provide a suitably reliable dataset if used on their own and so I could only derive results when used in combination with my post data.
I do not know if I should be grateful for this, as makes it more difficult for those using this algorithm to trust the results knowing that the input data is more limited than others which could force some to look to other resources to get to know me, but could also simply mean that I get removed from analyses entirely and so not even looked at at all?
More importantly though were the results that were found from the AMS and the quizzes that significantly varied from the results of my NEOPIR experiment, as shown below:
Where the value of 50 in this context is the average (and most common value) so to make it clearer, I created a graph showing the same results but displayed about the average so you can see what I am like compared to the general population:
Both the mapping of the results of my NEOPIR values to the Myers-Briggs, and from completing a dedicated (although I’ve never looked into it’s validity) quiz at https://www.16personalities.com – I have found myself consistently to be an ENTJ personality type. That is Extraverted, Intuitive, Thinking, Judging.
However my AMS results defined me as INTP personality type, that is to say Introverted, Intuitive, Thinking, and Perceiving.
Generally ENTJ corresponds to the OCEAN results I got from my self-reported questionnaires on everything but the three on taste in brands, music, and TV and movies. Additionally from reading descriptions corresponding to the two I would argue that I identify more with an ENTJ profile than an INTP, so I was surprised to see myself labelled most notably as introverted and wondered if this came as a result of the limited size of the dataset or the content?
ADHD And Social Media?
The first thing I wondered when seeing the mismatch between my perceived personality on social media and my perception was whether or not my ADHD traits were influencing these results?
I do not need to be an expert to know that the relationship between mental health and identity is a complex one – if personality can be influenced by your environment then it can be influenced by anything that affects how you perceive and interact with that environment. I may only have been formally diagnosed with ADHD as an adult but I have had it for my whole life and so even if the neuroscientists could determine the specific direct effects of ADHD in my brain’s functioning, I would never be able to go back over a lifetime’s worth of processes to figure out how my personality would have developed if those functions were perfectly average instead. Nor would much be gained by me from such an exercise as I could not go back in time and make it so, and even if my ADHD was somehow removed entirely tomorrow it would not remove the influence it has had already.
However, because those relationships are still not fully understood they are therefore unlikely to be accurately modelled in commercial algorithms. Yet it is undeniable that relationships do exist and so their absence is likely to cause inaccuracies between the results of these commercial models when trying to define an identity.
For example, a behaviour that could be attributed to my ADHD is when I am generally very inconsistent with my social media usage and don’t really find myself forming the typical “habit” usage behaviour that most websites and applications look to develop in their users. Instead of frequently spending several minutes at a time logged in, I can sometimes spend several hours straight using something intensely but then nothing for months.
This may have an effect on how my social media data is interpreted because for most people I think prolonged periods of absence from an application are seen as a retreat not from the platform itself, but from the content that the platform is presenting e.g. I am avoiding Facebook the website whereas others are avoiding the people and the media that Facebook is presenting them. Therefore this therefore could be a possible explanation of the introversion in my AMS personality profiles?
Hobbies, Interests, and Tastes
The Cambridge quizzes that formed personality profiles based on my music, brands, and TV and film opinions all rated me based on asking me questions about what were to me all very generic music, brands etc, that I was largely neutral in my opinion of. I would say that however I have a very widespread collection of interests as opposed to a few preferences developed in more depth, perhaps as a result of my ADHD or not. For example I will attend less than average number of events and sessions for each club and society I am a member of, but then I am a member of far more than average.
I believe that methods used to match personality based on things like music tastes work on the principle of homophily – roughly the idea that you are like other people who like the same things as you (read this good overview about how music data is handled like this for example). The problem with this is that if you like far more things of more variety than most people you cannot be all the corresponding associated personality traits simultaneously so I do not know how the algorithms determine results without any strong associations other than the variety suggesting high “openness” personality traits?
Those quizzes used an incredibly limited dataset however, so perhaps not very representative of what larger companies would be using when taking data from my assorted online viewings and Spotify. Yet even these, albeit larger, datasets are still limited due to the nature in which I use those services and I wonder if this is accommodated for in some way? For example, I am still a big fan of CDs and generally find vinyls cheaper from the charity shop for certain eras and genres of music so my Spotify usage is more for the music that I do not have, nor necessarily want to have, in these other formats such as Christmas or party playlists, or new music I have never yet listened to. I know that my Discover Weekly is always hilariously skewed to presenting some very cheesy tunes because it knows that I listen to tacky pop playlists when powering through one of my jobs cleaning on night shifts, but I still need to read about new artists or talk to people if I want to find things that make up the bulk of my music listening.
I know I could change this by converting all my listening time to being done through Spotify but say I don’t, what are the other impacts of this? What does this biased data mean for the information being derived from it? How does this affect the psychometrics and personality interpreted from it? This would depend of course on who, and how many, other users follow similar patterns to mine, and how similar they actually are to me, and how much other data was available about myself and them.
I am biologically female, and identify as a heterosexual woman and have always done so. I was therefore rather surprised when the AMS algorithms came back identifying me as male based on my Twitter usage?
However I am aware of the fact that my interests are often stereotypically classed as “male”, e.g. engineering and cricket, and I wonder if perhaps it is primarily these factors that are determining these results? What would be interesting to do would be to see what the results were if my nationality was changed, because the male dominance in my fields of interest are generally more significant in the UK than other countries.
I do not know whether this is something that would have great consequence on anything though, given that my gender should ideally not be relevant to any potential employers’ recruitment decisions, and so really only affects the marketing that is targeted at me.
The inaccuracies created by technological assessments of my identity are only of consequence if these are actually used to make decisions with. The majority of the decisions made day to day that affect me are those made by other people based on their intuitive perception of me, generally gradually becoming more accurate as they get to know me better.
However, the use of statistics and homophily in the algorithmic judgements are designed along very similar lines to the natural processes in the human brain, and the basis for the idea of subconscious biases.
I already know that it is very likely that I am subject to common subconscious biases in my ordinary everyday interactions as it stands because of experience, and knowledge simply that many features of my identity define me as being in the “tail end” of frequency spectrums i.e. statistically unlikely, and so more likely to be subject to these biases. For example I have high intelligence, I am a female in engineering, cricket, and other male dominated fields, and I have ADHD.
As a consequence of not fitting the default “normal” pattern expected by people I can flag up as a source of uncertainty, which for some makes me a novelty and others an assumed threat, until my specific patterns become known to those that invest the time to get to know them. I think that given that I have multiple identity features that cast me as a minority, I think it can be hard for others sometimes to disentangle which one(s) are causing these triggers?
Therefore I wonder if the results of the algorithmic judgements hold suggestions of how others may view me too? Arguably, their dataset should be larger than that of Facebook or Twitter, given that meeting me in person gives a lot of extra information such as appearance, body language, or tone of voice, on which to judge me. However, given the extraordinary dependency on social media for communication with so many people I know, I wonder how much this influences people’s perception of me through that filtering?
What are the consequences of this? If there is a mismatch between my social media profile and my actual identity does this trigger more of a perception of threat by creating more uncertainty?
There is a surprisingly large mismatch between how I define my identity and how it is calculated using a typical psychometric social media analysis processes, particularly extraversion, and the only metric I am consistently assessed correctly as is “Openness to new experience” in the OCEAN model.
The technological processes used to interpret the data are largely based upon homophily and probability which does not account well for features and patterns which are statistically unlikely, especially if they rely upon data that is complex simultaneously affecting multiple variables such as ADHD.
This could have a multitude of hypothetical effects on the users of those technological processes, but additionally could both affect and represent the natural subconscious biases used by individuals to form their judgements on my identity. Both the statistically unlikely nature of some of these features may trigger negative issues but this may be exaggerated by mismatch between my social media presence and my actual identity.
Could I do anything about this? Should I do anything about this? The answers are unclear at the moment and so for now, I do nothing but as I learn more I may, so watch this space!
My detailed results are on the next page if you are interested…