Many C-suiters used to eschew data-based insights in favor of “going with their guts” to make company-wide decisions. Gut-based leadership typically involved the decision-makers using their own backgrounds and on-the-job experiences to make guesses about what might be good or bad for the company. Importantly, this decision-making process was grounded in real-life knowledge acquired within jobs, companies and industries.
Given this history, it’s no wonder that making decisions based on such abstract entities as figures and statistics seems treacherous, especially given the speed at which data-accumulation and -dissemination technologies have proliferated over the past 20 years. Although the power of data-based inference has been evident for centuries, only recently have its applications grown widespread. To combat this perceived nascence and encourage decision-makers to embrace data-driven insights, analysts have to clearly convey the value of data-informed decision making. But how?
How Do I Use My Data?
Making Insights Clear
In order to take full advantage of data-based inference, practitioners must couch statistical insights in language that is relevant, understandable, and, most of all, contextualized for the target audience. The goal here is to use statistical findings to pique decision-makers’ curiosity.
When explaining how and why data address a specific problem, keep the central question asked of the data at the forefront of the discussion. If a part of the analysis is too complex to explain in the context of the problem, then it belongs in the appendix. This principle goes beyond leading with the results. Researchers must ground all facets of their analysis in the question’s context when communicating findings.
As an example, imagine the problem under consideration is that voluntary terminations have recently gone up at a major bank. Management suspects that these resignations could be a byproduct of potentially fraudulent sales practices. So, they want to investigate the scale of such untoward practices, if any.
Management could collect information by going out to individual members across the business or survey a sample of leaders in other industries about what they’re seeing in terms of turnover in their respective sectors. To get a quick, accurate picture of turnover using a wider lens, however, the best method would be to use human-resources data and statistical inference to piece together a narrative about what might be going on.
The best way to build this coherent narrative when explaining statistical results is for researchers to view their work as constructing a statistical case rather than looking for some root cause. Couching findings in terms of accumulating evidence for or against a conclusion also works much better than a presentation that silos the analysis into components (e.g. “Findings,” “Methods,” “Data Sources,” etc.) In working to build a statistical case, researchers automatically frame data outcomes in light of how and why they address the central question of interest to decision-makers.
Viewing data and statistical estimates as pieces of evidence in favor of or against some outcome serves to focus minds on the question underpinning their use. This runs counter to the uncontextualized “academic paper as presentation” style above that pits analytical results against the questions researchers are using them to address.
In the context of the accounts-fraud investigation, researchers might start with their main conclusions after stating the problem and question. Further, let’s assume that researchers did turn up statistical evidence of fraud. From this starting point, researchers present their findings as case exhibits with data, methods and visuals used to buttress each evidentiary plank in the overall conclusion fraudulent activity is the likely culprit for elevated quit rates.
For instance, presenting how quit rates this year were substantially elevated above their historical average would be a good first plank. From here, researchers could further highlight which lines of business were responsible for most of the increase and point toward potential incentives for unsavory business practices. For instance, if increases in personal accounts or retail banking roles were driving the overall trend relative to other lines of business, researchers might point to how new account generation being tied to bonuses or promotions creates an incentive for fraudulent activity. Presenting potential causes as they come up alongside statistical results helps to give credence to the overall point of each piece of statistical evidence’s usefulness.
What Else Could it Be?
Another useful method for contextualizing findings is to think about relevant counterfactuals. Bases of comparison for multiple possible answers engender critical assessment of potential solutions and get decision-makers thinking in terms of trade-offs. Simply ask what the most likely alternative that could have occurred is. This again helps to contextualize the full scope of the problem and implicitly shows data-based analyses’ primacy in determining what ultimately occurred this time and what is likely to occur in future.
Seeing how empirical evidence shifts the focus away from potential explanations of what has occurred and toward other ones serves to countenance their usefulness to decision-makers in parsing competing narratives. As decision-makers see the insights that statistics offer, highlighting other likely scenarios that did not occur reinforces statistics’ abilities to probabilistically rule out other likely explanations for the question at hand.
In our hypothetical example, then, alternative factors that might drive workers to terminate employment would be good counterfactuals to examine. One counterpoint could be that elevated quit rates are due to an aging workforce or correspond to young or short-tenured workers being more likely to quit sales roles. To explore this, analysts should look into average ages and tenures of quitting employees.
Another piece of evidence to examine would be whether quits fall disproportionately among lower-level workers versus workers in management roles. These patterns could also be juxtaposed with historical trends to fully flesh out the anomalous nature of why quits are occurring at such high rates in the specified lines of business.
Using data to shift the emphasis on to one hypothetical and away from others renders competing “business as usual” accounts moot. In this case, that shift was to fraudulent account activity and away from the competing alternative of terminations due to expected demographic patterns like retirements and typically high termination among new salespeople. Disproving alternative scenarios illustrates the powerful use of statistical evidence to weigh decisions in context.
So, once you’ve built a strong case for a likely solution and treated possible alternatives, you need to present the findings in a compelling way. A tempting method for quick explication is to use graphics and figures whenever possible. Numerical visualization has become hugely prominent over the past half-decade, especially given the accelerated rollout of new technologies that facilitate visualization. Recent examples include Tableau, Shiny, Power BI and others.
Although visualizing results can be critical for understanding and can help with counterfactual thinking, note that visualizations are not a magic bullet solution to the hard work of making statistical results relevant to decision-makers. Compelling graphics are critical for shedding light on data and estimated results, but they’re only a method of communication and not themselves communication. Another way of saying this is that captivating visuals are sometimes necessary to ensure business leaders use statistical findings to inform their decisions. The conclusions that visual evidence points towards, however, are more important. Appealing graphics and figures can be powerful means of communicating analytical findings, but they will not make researchers’ arguments for them.
Simply including visuals of time-series trends in inter-sector quit-rates to make the case to the bank’s brass that fraudulent account activity has occurred won’t be enough. A presentation of results that just consists of a hodge-podge of charts and graphics is chimerical. Their ordering, complementarity and coherence are all critical to reinforcing the overall findings that fraudulent-account activity is the most likely culprit for the yearly spike in voluntary terminations. It’s not enough to simply set up exhibits in court and rely on the judge and jury’s admiration of them to guide their verdict. You must make a case for them.
Leading With Data
Statistical results and the context generating them must guide potential solutions from researchers. If the problem’s likely culprit, in the example above, is perverse incentives, the environment in which the problem occurred must also be the environment in which the problem is solved. Simply firing and replacing fraudulent actors or shuttering specific locations where fraud has occurred are enticing solutions. They will not solve the fundamental problem, however. This is why the continued emphasis by researchers that statistical findings are the result of some process or processes generating them is so crucial.
This final point should serve to transmogrify data-based insights into this gut-level understanding. A coherent narrative undergirded by the weight of statistical evidence captivates, as well as informs, business stakeholders. In eliciting their curiosity, contextualizing statistical findings provides the relevance that decision-makers need to make informed decisions.