What Do You Mean? Interpreting Your Results

06.02.2017 |

Episode #8 of the course An introduction to data science by Roger Peng

 

In this lesson, we will illustrate several principles to help guide you when interpreting your results.

These principles are:

  1. Revisit your original question. This may seem like a flippant statement, but it is not uncommon for people to lose their way as they go through the process of exploratory analysis and formal modeling. This typically happens when a data analyst wanders too far off course pursuing an incidental finding that appears in the process of exploratory data analysis or formal modeling. Then the final model(s) provide an answer to another question that popped up during the analyses rather than the original question. Revisiting your question also provides a framework for interpreting your results because you can remind yourself of the type of question that you asked.

  2. Start with the primary statistical model to get your bearings and focus on the nature of the result rather than on a binary assessment of the result (e.g., statistically significant or not). The nature of the result includes three characteristics: directionality, magnitude, and uncertainty. Uncertainty is an assessment of how likely the result was obtained by chance. A great deal of information that is required for interpreting your results will be missed if you zoom in on a single feature of your result so that you either ignore or gloss over other important information provided by the model. Although your interpretation isn’t complete until you consider the results in totality, it is often most helpful to first focus on interpreting the results of the model that you believe best answers your question and reflects (or “fits”) your data, which is your primary model. Don’t spend a lot of time worrying about which single model to start with, because in the end you will consider all of your results, and this initial interpretation exercise serves to orient you and provide a framework for your final interpretation.

  3. Develop an overall interpretation based on (a) the totality of your analysis and (b) the context of what is already known about the subject matter. The interpretation of the results from your primary model serves to set the expectation for your overall interpretation when you consider all of your analyses. Recall that this primary model was constructed after gathering information through exploratory analyses and that you may have refined this model when you were going through the process of interpreting its results by evaluating the directionality, magnitude, and uncertainty of the model’s results. External information is both general knowledge that you or your team members have about the topic, results from similar analyses, and information about the target population.

  4. Consider the implications, which will guide you in determining what action(s), if any, should be taken as a result of the answer to your question. After all, the point of doing an analysis is usually to inform a decision or take an action. Sometimes the implications are straightforward, but other times the implications take some thought. An example of a straightforward implication is if you performed an analysis to determine if purchasing ads increased sales, and if so, did the investment in ads result in a net profit. You may learn that either there was a net profit or not, and if there were a net profit, this finding would support continuing the ads.

 

Tomorrow, you’ll learn how to best communicate the results of your data analysis.

 

Recommended book

“Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die” by Eric Siegel

 

Share with friends