The 2016 United States Presidential election will go down in history for many reasons, one of which is the failure of the vast majority of polls to predict the winner. The USC / LA Times Daybreak poll was the only major statistical predictor of a Trump victory. Why is that?
The answer to that question is important to manufacturers because of the expanding role of data in our businesses. If all those professionals can get it wrong, how can we ensure our data and analysis are valid?
While I have a solid quantitative background, I don’t have enough information about the leading surveys to say why they got it so wrong. Questions can be biased, sampling techniques can be biased, and good statisticians can use the same data to draw different conclusions. But there was one major difference in the LA Times survey: it included questions about the intensity of voter preferences for one candidate over others. The others generally asked the black or white question of “whom would you vote for?”
Your company may be surveying customers, suppliers and employees at least annually. Is that data giving you information that accurately reflects the thoughts of those groups?
Manufacturers commonly report order – to – ship lead-times. If an order is on hold while awaiting credit approval, is that wait time part of the reported lead-time? Does “ship time” mean when it’s on a truck, on the dock waiting, or when an invoice is generated? The answer may be a function more of software configuration than business need.
As machines self-report uptime and downtime, does the data know when the machine should have been running? Uptime may indicate running unneeded product or larger lot sizes. Downtime may mean a company meeting. Is your data collection valid for formulating important business decisions?
ERP, MES, Barcoding, RFID, IoT – the list of data-oriented investments the average manufacturer is making goes on and on. But how do you know the most appropriate data is being collected and analyzed? One client recently assigned a new manufacturing engineer to perform time studies on machine center setups. That study was incredibly detailed, but very misleading to the company leader using the data for a decision. Had the ME known what question the data would be used to consider, he would have collected and presented the data differently.
Let the 2016 election provide valuable insights to your organization. The data can be true, but wrong. The analysis can be valid, but misleading. I strongly encourage every manufacturer to blend strong statistical skills with business acumen. And if the data shows something surprising, it may just be right!
As published in AME’s Target Online
