Data Analytics Part 2 – false insights

even with the most advanced analytics decision-making can be undermined by false insights there are three big traps which must be avoided Keigo means garbage in garbage out it stands to reason if you feed bad data into your analytics you're going to get some highly misleading outputs this sounds like an obvious statement but there are so many ways that data can lose its validity and become a liability and here is the problem data is data there is no way of knowing if data is good or bad simply by looking at it for example if you read from a table of information to calculate how hard and long to deploy thrusters on a spacecraft you would hope that data is accurate this was the case when international teams collaborated on the Mars climate orbiter project for NASA which launched in 1998 unfortunately one of the teams assumed NASA was working in imperial units when in fact they had moved to metric some years earlier the result was a hundred and twenty five million dollar project finishing in disaster as the craft missed its orbit and was lost forever it goes to show that if rocket scientists can get this wrong so can businesses so rule number one check your data this is achieved by getting your data infrastructure in order and assuring the data with strong governance processes this problem is made much worse by the second trap people see what they want to see in psychology this is called confirmation bias the problem here is that people are naturally programmed to filter their experience of the world based on their feelings beliefs and past experience that NASA scientists in the orbiter example would have seen validator because they believed in the expertise of those supplying it here is another example from the world of business imagine that you just joined a new company and you hear this statement from your CEO at your first allstaff business update meeting last quarter we saw record-breaking sales with one hundred and fifty percent year-on-year growth this is the picture of a company showing some great results and you are likely to believe it because it validates your choice to join the company however if you had been there longer you may also know that the same quarter a year ago was horrendous with sales performance hugely below target based on exactly the same figures a very different perspective could be described last quarter we had a single big deal which saved our bacon at least the overall result was better than the disaster last year so 150 percent growth can be good or bad depending on how the business did last year and indeed what the targets look like for this year hence rule number two look past the headlines data must be viewed objectively fully and in context before drawing conclusions what else could possibly go wrong as Mark Twain is often quoted as saying there are lies damned lies and statistics ironically he never did say this regardless the expression captures nicely the ability to use numbers to reinforce any argument you wish to make if you look hard enough in data there are ways to justify your beliefs this can be a problem even with the highest personal integrity and honesty because numbers can be deceiving for example let's say I have a theory that the rising consumption of cheese is causing a corresponding increase in deaths I may believe it's likely based on the popularity of cheese and knowledge of the rise in obesity if I were to look hard enough I might find this data however this is not the whole picture I am missing a crucial factor about the nature of these deaths I need to dig deeper and get the full context these deaths are in fact due to entangling in bedclothes there clearly is a strong correlation between these two factors but actually no evidence that one causes the other correlation does not equal causation in fact these are coincidence if you look hard enough you can find highly misleading evidence in numbers the correlation trap is just one example in statistics this is called overfitting Henschel number three data first insights should follow not vice-versa if you look too hard for data which supports a hypothesis you are likely to find it whether it's right or wrong you

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *