suicide attempts around Valentine's Day than Christmas Day. I didn't
know that and I think many other people didn't know that either.
It's an unexpected correlation. There are very few other studies in the
area. It wasn't something which was deemed to be worth investigating so
no one posed the research question. These are what drive research in the
sense that someone, somewhere decides that something needs to be
investigated then expends the effort to find out if it's true.
The paper below is a very simple one. The data analysis is easily
achieved and there's no complexity in the methods. It was at little to
no cost because it's an observational study rather than a true
experiment. There's no qualitative component to investigate why people
attempted suicide around Valentine's Day. The authors have made an
assumption that it's because of unrequited love.
There are probably many other unexpected correlations that haven't been
discovered because no one's thought to properly investigate it. This can
be because people assume something is true, for example that there's a
link between the moon and madness, but it's only till there's research
evidence obtained that this can be proved to be true. When the reviews
of lunacy are done it turns out that there's no link between the mood
and madness in high quality studies. It 's the same for men and the
relationship between help-seeking and suicide.
There are many of these undiscovered truths which remain undiscovered or
falsely assumed because there's no research done, no academic or doctor
thinks there's a question which needs to be answered.
I consulted on the IAPT research program. IAPT represents a significant
research opportunity. Not only is it a very well funded program the
Improved Access to Psychological Therapies scheme also has a huge system
of measures integrated into the system. Evidence in mental healthcare is
like an Essex girl's underwear: rarely seen where it should be.
Psychological therapies research is often dogged by tiny sample sizes.
IAPT presents an opportunity to investigate with a new degree of
accuracy the potential of psychological therapies in mental health
treatment.
One of the research questions was about therapist-client effects. These
have been investigated in studies going back to the 1970s and seek to
understand what effects other than the mode of therapy happen between
the therapist and client to make for better therapy.
I did my usual thing of criticise the research because it may not
influence commissioning. Whatever was found to be the significant factor
may not be something which the NHS could ensure patients could get or be
something they could use as requirement for therapists.
There's also another problem: the problem of the limitations of the data
set. It's not an experiment so it has to use what's already there: the
IAPT outcome data. This records some characteristics but not all, for
example it might record gender or religion but, as I suggested, it might
not have attractiveness (and hard quality to measure objectively but one
which we can all recognise subjectively).
I made the point to the research team that an attractive therapist might
offer the best outcomes. It is based on a guess that people who are
attractive are listened to more and people want to spend more time with.
It is perhaps a bit cynical of the human condition but also probably
accurate.
It happens to be that I guessed this as a factor but I don 't think
there's any research into the influence of a therapist's personality and
beauty. I'm just guessing from personal experience that I like chatting
to attractive people and want to see them more. They may not be as
interesting oir 'sound' as unattractive people who've had to develop
other qualities, perhaps because they lacked beauty (or the confidence
about their beauty).
This might make me sound like a lecherous git or just someone who's
honest about the human condition and has no airs about it.
The research team obviously hadn't considered this factor. They probably
thought it was irrelevant because there's no way the NHS could
commission more attractive therapists. I think I made my point well.
But there was another aspect too. It was the software they were using.
It seemed like what they'd need to is specify a set of criteria to
examine, for example age difference relationships or gender, and
manually run the calculations for each variable.
I studied electronic engineering, not psychology or anything like that.
This is fucking ludicrous shit to me. Having to manually run each
variable. How pre-millennium. This is just pants.
It is possible for a computer program to run all the variables and
permutations of variables then output the ones with the highest
significance and effect. It might be that black men who are 18-35, right
handed and shave every day are best treated by Asian women who're
divorced and wear makeup. This is a silly example to make the point that
the effects are unknown and their interplay is unknown. What's available
is a dataset with lots of variables (obviously not some of the ones I
included in the silly example). A computer could run calculations for
the combinations of variables then compare them automatically before
outputting the variables with the highest statistical significance and
effect size.
I might not have understood their software but I don't think it could do
this. It was essential like Excel on steroids so could do quantitive
modelling (make graphs) at a very advanced level but it couldn't
automatically run all the possible variables available in the dataset.
It is a lot of number crunching but not for a computer. Modern
processing power is very cheap. The performance of supercomputers
pre-1990 are now matched by home PCs.
The software could work out what was the biggest effect and the one
which worked for most people (or a subgroup).
The reason this is valuable is because there is the problem of
expectations. Our mind's are limited and can be biased. Computers
aren't. They're dumb and do what they're told very efficiently. They
don't exclude options from investigation because they expect they won't
have any effect. They'll just get on and run the data.
This sort of intelligent computing is not inconceivable. In fact I'm
surprised there's not a commerical product which can do this. It's just
a step more than the software the IAPT researchers were using. This may
be quantiative mining and modelling analysis automated to run
unintelligently but by removing the human factor may be able to find new
facts or dispel mistruths.
Obviously as the number of variables increases the computational
requirements increase exponentially but Moore's Law continues to offer
cheap computing power. Distributed computing such as the system which
powers Google or the SETI@home project are already using the power of
desktop computing to replace supercomputers. Computers which used to
cost tens of millions are being replaced by the spare processing power
of home computers in the SETI@Home (and other @Home projects) so the
search for extra-intelligence or the data from the Large Hadron Collider
can be processed quicker and cheaper.
As more data is made available electronically the potential for this
sort of computing application are very high. It's something which seeks
correlations. I think there's already been more sophisticated programs
written and experimented with but this one seems like it'd work in the
real world.
No comments:
Post a Comment