“I have always said if you torture statistics long enough they will eventually confess.”
– Alan Simpson, former Senator from Wyoming, in an August 13 2012 interview with Deirdre Bolton on BloombergTV’s In The Loop, video embedded and excerpts transcribed at Tyler Durden’s August 13 2012 post on Zero Hedge titled “Alan Simpson Confirms Reality: ‘All The Things You Love Will Not Come To Pass'”
First off — I actually like statistics.
I will also readily admit statistics is not a field that has set answers, rather it is a field where the answer you get will depend greatly upon what question you ask and how you ask it.
And don’t even get me started on whether or not your sample is large enough or diverse enough to be a truly representative sample (in which case you can use some types of statistical methods and assumptions) or not (in which case you have to make other assumptions and possibly use other methods too). Ugh.
But to most people, statistics sounds and looks like regular math.
And most people are used to regular math having firm answers, like 6 x 7 = 42. There’s even math proofs you have to go through in high school algebra to drill into you that that 6 x 7 = 42 no matter how you ask the question. 6 x 7 = 7 x 6 = (7) x (6) = (7 x 6) = (((6) x (7))) = 42!
I think most statisticians are so used to dealing with statistics that they don’t realize a lot of people don’t like math, have never liked math, and are not so eager to challenge your huge equations that they’re willing to relive the unpleasant memories from high school algebra class.
Which leads a lot of statisticians (and people who churn through huge amounts of statistics, like economists) to assume that the reason nobody is challenging their work is because everyone is thinking “Hey! I think they got the right answer!”, instead of the more likely truth “Ack! I got lost somewhere between the third parentheses, the fourth exponent, and the second capital sigma sign. I’ll just nod my head and hope to God they don’t try and explain it again.”
So, yes, statistics do need to be tortured. After you’ve gone through six sets of assumptions, three different methods, gotten about twelve different answers, then maybe, just maybe, you might start to get an idea of what the situation truly is. (In one set of data I tried to analyze at one of my jobs, the situation was “Wow, that series of tests didn’t prove anything as far as I can see” — and when I presented my findings to the rest of my engineering team and our team leader, they also all looked and said “Nope, no discernible results that we can see either.” Which, believe it or not, is actually a valid result since it means there was no clear correlation in what I was testing and that went a long ways to discouraging any assumption of causation.)
And no, I haven’t watched the video yet. Given the transcriptions at the Zero Hedge article, it looks to be well worth watching. Hopefully I’ll watch it in the next few days.