Apdex is a way to study measurements of any experience that can be interpreted on a scale ranging from excellent to unacceptable. It can be used to rate everything from application response time, to food quality, surgical outcomes, time to repair, bandwidth delivered by an ISP, and time to pour a beer. The sky is the limit. Armed with measurements of many “events,” it is a way to summarize outcomes with a simple, easy-to-understand score.
Many statistical data analysis methods are based on a normal distribution of values (the familiar bell curve). There are three problems with this common approach. First, it assumes that the dominant values are most important (mean or median). Second, it assumes that values further from the mean or median are less important (deviations from the norm). Finally—and most importantly—it assumes that values on either side of the dominant value have an equal impact (e.g., it assumes that 10% too high is just as bad as 10% too low, which often is not true).
Apdex enables you to present a set of measurements so that a few, significant results receive more weight in the outcome. It also gives a voice to the group of experiences that typical statistical approaches dismiss as outliers.
Furthermore, the three “buckets” in the equation need not be centered on the mean. For example, let’s assume that the objective is to study the accuracy of blood pressure tests performed at a number of health clinics. The hypothetical question is—how often does a clinic make a misdiagnosis that leads to unnecessary and potentially costly treatment?
In our sample study, we send 100 certified healthy individuals into each clinic for a blood pressure test. All of the subjects have normal blood pressure. Some clinics note, incorrectly, that a patient’s blood pressure is too high or too low. In this situation, standard deviation is unhelpful. If a blood pressure result is too high, the clinic will likely prescribe treatment, and if it is too low, the clinic will likely tell the patient to return later.
The point is, there is a material difference between blood pressure results that are out of the normal range depending on whether they are high or low. Most statistical methods would treat a positive or negative deviation from normal equally. Apdex lets the analyst assign the low blood pressure patients to the “tolerating” group and the high blood pressure results to the “frustrated” group. The Apdex scores for each clinic in this experiment will identify the clinics that are inappropriately adding to the cost of patient care.
Traditional use of the Apdex formula when applied to application response times as seen by the user:
Apdex = (Satisfactory samples + 0.5xTolerating samples + 0xFrustrated samples)/Total samples
New use of the Apdex formula when applied to blood pressure accuracy influence on health care cost:
Apdex = (Normal samples + 0.5xLow samples + 0xHigh samples)/Total samples
Note: This post is not intended to provide health advice. See a healthcare professional regarding blood pressure.