3 Analytics Terms for Success | npENGAGE

3 Analytics Terms for Success

By on Apr 24, 2012


We often use history to predict the future. I checked wunderground.com’s history of temperature averages before selecting my summer vacation destination. Predictive analytics has become a hot-button term over the past two years as business intelligence vendors have begun to incorporate simulation and forecasting into their offers. In order to predict with greater confidence, here are three terms you must know.


It is easy enough for the human eye to observe, “It looks like the test significantly outperformed the control.” But statistical measures are necessary to interpret the repeatability of the observed result. Statistically significant findings should be reported with a confidence level. Think of a 95% confidence interval as an indication that if you were to take 100 different samples from the same population, the test would “outperform” the control in about 95 of the samples.


Did you note above that “significance” requires your sample observations to be representative of the greater population? For example, Congressional representatives are supposed to be a sample of the population within their districts. Ironically, Congress is seldom a good example of a truly representative sample. Scrutinize the universe under observation. Very likely you are making observations today that will influence your strategy implementation tomorrow. You cannot control for environmental changes over time, but as much as possible you should manage your test sample to represent the population in your future. Read more about Tests, Controls and Results.


Be on the watch for confounding. Hidden variables that are correlated with both your dependent variable and your independent variable(s) are called confounding variables. The classic example is ice cream sales as a predictor of drowning deaths. There is correlation, but the underlying influencer of both is temperature and season encouraging both ice cream consumption and water sports. Avoid this oversight by brainstorming potential hidden variables with your colleagues.

Keep these three concepts in mind when you are creating your own predictive analytics hypothesis or reviewing analytics provided by others. When you know your level of significance, know that your sample is representative, and have accounted for hidden variables you will be able to support strategic decisions with confidence.


Rebecca Sundquist is a lead analyst in Convio’s strategy practice group, her goal is to lend confidence to strategic decisions. She investigates constituent data to uncover trends and confirm or deny hypotheses. Rebecca practices the art of info viz (information visualization) with a commitment to simplicity. She wants to help clients see clearly where they have been and how to reach their goals.

Since 2004, Rebecca has worked with nonprofit data, implementing and monitoring acquisition, renewal and reactivation practices. She also has experience with donor upgrade strategies, activity cross over and sustainer programs.


From time to time, a guest blogger will appear on npENGAGE. Guest bloggers are industry experts contributing useful, relevant content to the conversation on npENGAGE. If you are interested in being a guest blogger, contact the editor.

Leave a Reply

Your email address will not be published. Required fields are marked *