Analytics Magazine Analytics Magazine, September/October 2014 | Page 32

My th s of A n a ly t i c s than seven times after adjusting for inflation. In lockstep, deaths due to heart disease more than halved, making it easy to conclude that the two are closely related. Medical care advances are believed to have contributed to lower heart diseaserelated deaths through improved diagnosis and treatment. However, general lifestyle and diet changes also played significant roles. (Read more.) Furthermore, a review of other chronic disease trends reveals that some medical conditions, such as diabetes, have worsened (see Figure 1). There’s also myriad other factors potentially related to escalating healthcare costs, such as an aging population, greater administrative expenses and broader marketing pressures. For another take on how correlations potentially mask deeper relationships, check out Christopher Knittel and Aaron Smith’s paper “Ethanol Production And Gasoline Prices: A Spurious Correlation.” Takeaway: Seriously consider whether correlation truly reflects a relationship or simply masks the influence of one or more hidden intervening variables. 4. Random sampling ensures representation Unless you’re working with full universe coverage, some form of sampling is usually required. And while fantastic 32 | a n a ly t i c s - m a g a z i n e . o r g in theory, true randomness is difficult to achieve. Transactional data are constrained by membership and/or opt-outs; surveys face non-response; and social media content is subject to issues related to self-reporting. Beyond sampling, the challenge of unbiased representation is exacerbated by a number of factors ranging from human predispositions to herd mentality. Sinan Aral wrote a convincing piece in the MIT Sloan Management Review, for example, explaining the tendency for online customer reviews to be abnormally j-shaped rather than bell-curved. Referencing different studies, Aral explains how herd mentality can lead to a disproportionate concentration of positive ratings skewing online reviews over short and long terms. (Read more.) It’s another example of how things aren’t always what they appear to be. Takeaway: Accurate representation without some form of post hoc control is frequently illusive. 5. People are rational Humans act irrationally. As consumers, we often derive greater satisfaction from the same item if it costs more (not less); we let decoy options cause us to make suboptimal decisions (such as buying something bigger than we normally would); and we frequently stick w w w. i n f o r m s . o r g