Pub. 4 2016 Issue 3

www.uba.org 12 6 Biggest Housing Data Mistakes H ave you ever listened to a newscast where it became painful- ly obvious that the reporter had absolutely no real knowl- edge of the subject matter? Unfortunately this is particu- larly common with news involving data. In our fast-paced world where only the most recent data is reported in a rushed headline, we as data consumers can often draw skewed, if not completely wrong, conclusions from the announcement. The purpose of this white paper is to briefly delve into six of the most common errors when it comes to data reporting or data con- suming. If you’ve ever had a time when housing numbers didn’t seem to make sense, or if you have colleagues who are not used to digesting this type of information, our hope is that this docu- ment helps clear up some confusion. While far from an exhaus- tive list, we see these particular issues come up very frequently. #1 ‘Tis the Season(ality) This section isn’t about the holidays, at least not exactly. Housing data is frequently reported without an indication of whether it is seasonally adjusted or non-seasonally adjusted. What’s the differ- ence? Seasonally adjusted data is a statistical creation designed to cancel out the natural ebb and flow of data activity in different times of the year, while non-seasonally adjusted data is more raw. Seasonal adjustment typically translates a single month’s number into an annual rate. So for instance, when the Census Bureau reported that February 2015 new home sales came in at 539,000, that does not mean 539,000 new homes were sold in February (those of us in the housing industry would be very excited were that the case!), it means that based on the pace of activity in February compared to the same month in years past, we’d wind up with 539,000 new home sales in 2015 if every other month performed as well respective to their historical counterparts. If that sounds a bit confusing, it is. But the benefits to seasonal adjustment is that we can compare one month to the next, so we can judge seasonally adjusted numbers from February next to seasonally adjusted numbers from January, and have a valid com- parison. We cannot do that with non-seasonally adjusted figures, because in this case February new home sales are almost always higher than January new home sales (it’s only happened twice since 1990 that January posted a higher number), so it doesn’t tell us much. Instead, if you want to use non-seasonally adjusted numbers, you need to compare them to the same month a year (or more) ago. So seasonal adjustment can be helpful for looking at things month-to-month, but it generally brings higher margins of error with it, which we’ll discuss later. #2 Self Confidence Now that we’ve talked about seasonal adjustment, we can talk about something that often comes with it: higher margins of error. With many of the government data releases like those from the Census, you’ll also see a “90% confidence interval.” What does that mean to us as data consumers? Surprisingly, it means that often the change in direction described (sales are up, sales are down) is actually not statistically significant. In other words, because of either sampling errors (naturally with any survey you have some margin of error when applying the results to everyone) or non-sampling errors (bias, nonreporting, under- coverage, etc.), often times the agency releasing the information doesn’t really know if the data went up or down, despite the fact that the headline is reported as if they do. Since we’re discuss- ing the US Census, they state this pretty clearly on their releas- es, “90% confidence interval includes zero. The Census Bureau does not have sufficient statistical evidence to conclude that the actual change is different from zero.” “Well,” you think, “that can’t happen that often, right?” It occurs more often than you would probably think. With our example of new home sales above, turns out that the seasonally adjusted percentage change rate from January to February had a 90% confidence window that included zero. In other words, they don’t really know if February was better or not. Looking back at all of 2014, this was the case in 9 out of the 12 months. You might want to read that sentence one more time, because it means in 2014 the Census didn’t know if New Home Sales went up or down 75% of the time. Yet that is rarely, if ever, discussed when the data is released. This happens in many different data releases, not just New Home Sales, so always be sure to figure out just how much confidence they really have in their numbers. #3 Wait, Wait, I Take it Back! So now we know to look at whether something is seasonally adjusted or not, and whether it is actually changing like the