Safety Data: Signal and Noise Continued from page 51 range of the data. Often, this helps us understand variation in the data set. Inferential statistics, instead of simply describing variation, attempt to explain that variation as result- ing from the influence of one or more variables in the system. As an example, suppose we wanted to describe instances of recurrent training lapses within our workforce. We might find that the mean lapse in training currency across line techni- cians is 23 days, but ranges from zero to 112 days past due. While this doesn’t tell us much, it does indicate a wide variation in training lapses, whereas a range from zero to 13 days past due with a mean of seven days indicates much tighter variation within the data set. Inferential sta- tistics extend the impact of the data by allowing us to look beyond con- clusions about the immediate data. Correlation between observations bridges the gap between descriptive and inferential statistics by math- ematically describing relationships Aviation is nothing if not a lot of variables all changing at the same time! Aviation Business Journal | 3rd Quarter 2017 between variables. We might choose to examine whether, during times of the year where training lapses are high, ramp hazard report frequency also increases, or ramp observa- tions indicate issues with SOPs. We’ve looked primarily at cor- relation here because it is integral to most common statistical tools, but it is far from the end of what statistics can do for us. Inferential statistics have many uses in aviation aimed at examining how multiple variables interact to affect something of interest that has yet to occur. Sound familiar? Of course it does. Aviation is nothing if not a lot of variables all changing at the same time! While this article just scrapes the surface, you can begin to see how elements of the system can be tied together logically to supplement our gut feelings about the nature of how work gets done, and how we can improve safety and effectiveness. Both descriptive and inferen- tial methods are heavily focused on quantitative data because it is generally numeric and more consis- tently manipulated. Qualitative data analysis seeks to identify themes and patterns within categorical or plain-text information. Even with- out specific statistical operations, qualitative information can provide meaning and context to the quantita- tive analyses. Correlation, the axiom goes, does not imply causation. Sure enough, that’s often the case, but strong correlation (one variable is interdependent with another) is good reason to start asking further ques- tions and using qualitative data to find out more. Qualitative tools often revolve around assessing reliability among evaluators or raters, or may even extend into data mining and text analysis tools (often referred to as big data analysis). More often than not, there is a benefit to mixing methods, and to using qualitative data to give deeper meaning to the quantitative analysis, helping the data tell a more complete story. Data Sharing Although all of these tools for data aggregation and analysis can function within a single organization, the real power of statistical analysis is in iden- tifying and unpacking trends across an industry or segment. In airline operations, safety data is aggregated and analyzed as part of the Aviation Safety Action Program (ASAP), as well as through outsourced Flight Operational Quality Assurance (FOQA) databases. Data is used by air carriers around the world to dynami- cally modify training (using Advanced Qualification Programs) and evaluate operational performance (through Line Operational Safety Audits). The good news is that these programs are scalable to meet the needs of small operators, especially with the cooperative power of organizations like NATA to act as a clearinghouse, or at least as a coordinating point, for data analysis in the absence of participation in programs like the FAA’s Aviation Safety Information Analysis and Sharing (ASIAS) for many segments of general aviation. Learning from data to increase SOP effectiveness, capitalize on useful safety methods, prevent injuries, and understand normal work are all bene- fits of working toward more collabor- ative industry data analysis; however, Continued on page 55 53
