1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

People named Mohammed apparently pay more for car insurance

Discussion in 'General Discussion' started by dowie, 22 Jan 2018.

  1. Amp34

    Caporegime

    Joined: 25 Jul 2005

    Posts: 28,855

    Location: Canada

    Probably not many. You can only drive on a foreign licence for up to 12 months and then (if outside the EEA or a citizen of a reciprocal country*) would need to pass a driving test to get one.

    You can't just come from the middle east (for example) and get a UK licence without first passing the standard driving test.


    *The list of countries with which Britain has reciprocal agreements includes: Australia, Barbados, British Virgin Islands, Canada, Falkland Islands, Faroe Islands, Gibraltar, Hong Kong, Japan, Monaco, New Zealand, Republic of Korea, Singapore, South Africa, Switzerland and Zimbabwe.
    https://www.expatica.com/uk/moving-to/Exchanging-your-drivers-licence-in-the-UK_107710.html
     
  2. Tunney

    Capodecina

    Joined: 11 Oct 2004

    Posts: 14,549

    Location: London

    Hard raw data that’s fed into an algorithm written by a human being. A human being with biases, both conscious and subconscious.

    Data is nothing without interpretation.
     
  3. TNGL

    Mobster

    Joined: 2 Jul 2010

    Posts: 3,095

    This isn't the first time something like this has happened:
    https://www.theguardian.com/busines...ay-ethnic-penalty-car-insurance-report-claims

    I would be very curious to know what sort of feature engineering went on at the insurers listed in the OP. As for the algorithms having some sort of bias, I disagree, however it is possible for human bias to be introduced into the data that the algorithm will 'learn from'. Either way, I think the wally that insisted on incorporating names as a rating factor should perhaps carry out a more in-depth analysis into the potential causal relationship between the two.

    The logic might have been <take name> --> <classify ethnicity and/or religion> --> <use ethnicity and/or religion as a way to rate policyholder> ---> adjust premium upwards/downards... At least that way it's less obvious you are discriminating against them. I don't suppose anybody has tried names such as Hussain, Islam, Safwan etc to see if they get inflated premiums?
     
  4. Tunney

    Capodecina

    Joined: 11 Oct 2004

    Posts: 14,549

    Location: London

    There’s been some research into whether the algorithms used around us exhibit bias and the conclusion is a clear ‘yes’.
     
  5. kindai

    Sgarrista

    Joined: 9 Aug 2013

    Posts: 7,937

    Location: Bromsgrove

    :rolleyes:

    Im sure when they programmed it, they said, look at name vs accident rate oh, and dont forget if named mohammed add 10% because discrimination?

    Really?
     
  6. TNGL

    Mobster

    Joined: 2 Jul 2010

    Posts: 3,095

    I just read the article and two of the articles it references around gender biased nlp and teacher ranking, both of which heavily point to the data and not the algorithm. Of course models can be tuned to exhibit some sort of bias but the very generic algorithms off which a lot of machine learning is based aren't biased at all. In fact, how could you even go about adding a bias to say, an SVM?

    On the other hand, it would be very easy for me to manipulate my dataset whereby I increase all occurrences of claims by people called Muhammed by 30% and artificially increase the settlement value of these claims by 400%.

    GIGO, garbage in, garbage out.
     
  7. explicit4u

    Capodecina

    Joined: 29 Nov 2008

    Posts: 11,280

    Location: London

    Wouldn't the algorithm just replicate bias from bad data?
     
  8. Greebo

    Caporegime

    Joined: 20 Jan 2005

    Posts: 41,619

    Location: Co Durham

    Just slightly changing your job title can change your premium by £1000 anyway.
     
  9. Jono

    Man of Honour

    Joined: 18 Oct 2002

    Posts: 13,261

    Location: Northallerton/Harrogate

    With the exception of say Lewis Hamilton... People from ethnic minority backgrounds are generally much worse drivers than anyone else, at least around where I usually drive.
     
  10. Stretch

    Capodecina

    Joined: 14 Feb 2004

    Posts: 13,507

    Location: Peoples Republic of Histonia, Cambridge

    The quest for profit is generally colour blind.
     
  11. Em3bbs

    Soldato

    Joined: 26 Dec 2011

    Posts: 5,478

    Location: City of London

    Wow, so a single human being wrote the whole algo for all these insurance companies? They must be hella busy working in between EDL marches and the like.
     
    Last edited: 23 Jan 2018
  12. Skidder

    Man of Honour

    Joined: 28 Nov 2007

    Posts: 12,650

    Profit Mohammed?
     
  13. [email protected]

    Capodecina

    Joined: 30 Nov 2005

    Posts: 11,857

    Campaign for free insurance for all Muslims, we must not allow this ministry group to financially victimised.


    Can I change my legal name to Mohamed to get cheaper insurance, if we are successful?
     
  14. gumpy321

    Hitman

    Joined: 15 Jan 2011

    Posts: 611

    Location: Manchester

    Very good.
     
  15. Tunney

    Capodecina

    Joined: 11 Oct 2004

    Posts: 14,549

    Location: London

    It’s unlikely that it was a single person but you don’t have to be an EDL supporter to write a biased algorithm or seed a machine learning system with bad data.

    :D
     
  16. Le Clandestin Brun

    Sgarrista

    Joined: 6 Sep 2007

    Posts: 9,043

    As a person who has grown up with an Indian name I can say we've all known this to be the case since we first started driving. It's the way of the world (of insurance).
     
  17. Orionaut

    Sgarrista

    Joined: 2 Aug 2012

    Posts: 7,813

    But Insco's dont CARE about why Mohamed has a higher risk, they dont care why somebody involved in a non-fault accident has a higher risk either. All they care about is that they are, and charge accordingly.

    I imagine Inscos have all sorts of data sets that are really pretty controversial, whether they use them all or not I dont know, but I would be fascinated to see them...
     
  18. Dis86

    Suspended

    Joined: 23 Dec 2011

    Posts: 28,576

    Location: Northern England

    So the possibility that it's perfectly good data with a solid statistical background isn't a possibility then? Or probability for that matter?
     
  19. robfosters

    Caporegime

    Joined: 1 Dec 2010

    Posts: 40,909

    Location: Welling, London

    True. Not £1000, but when my brother qualified as a gas engineer, removing the apprentice from his job title dropped his premium by £200. I can only assume apprentices have more accidents and are viewed as younger and higher risk than fully qualified people.
     
  20. Tunney

    Capodecina

    Joined: 11 Oct 2004

    Posts: 14,549

    Location: London

    The point I’m making is that it’s a fallacy to believe that anything based on “raw hard data” is free from human error and bias.