Analytical and Descriptive Statistics
Descriptive statistics summarize the data you actually collected, using numbers such as the mean, median, range and standard deviation, plus charts like histograms. Analytical statistics (also called inferential statistics) use a sample to draw conclusions about the larger population or the process that produced it, using confidence intervals and hypothesis tests. In Six Sigma you describe the data first, then analyze it.
There are two broad types of statistics. Descriptive (sometimes called enumerative) statistics answer "what does this data look like?" Analytical statistics answer "what can I conclude beyond this data, and how sure can I be?" Both rely on the same sample. The difference is the question you ask of it.
What is the difference between descriptive and analytical statistics?
| Descriptive (enumerative) statistics | Analytical (inferential) statistics | |
|---|---|---|
| Question | What does the data we have look like? | What can we conclude about the population or process, and how confident are we? |
| Scope | Only the data collected | Goes beyond the sample to the population or future output |
| Typical tools | Mean, median, mode, range, standard deviation, percentiles, histograms, Pareto charts, dot and box plots | Confidence intervals, hypothesis tests, ANOVA, regression, capability estimates |
| Uncertainty | None stated: the numbers are exact for this data | Always stated: confidence level, p-value, margin of error |
| Depends on | Accurate data | Accurate data plus a representative sample and valid assumptions (for example, a normal distribution or a stable process) |
| Main DMAIC phase | Measure | Analyze and Improve |
Common descriptive statistics
- Center: the mean (arithmetic average), the median (middle value when sorted) and the mode (most frequent value). The median is less affected by a few extreme values, which matters for skewed data such as cycle times.
- Spread: the range (largest minus smallest), the standard deviation (typical distance from the mean) and the variance (standard deviation squared). Six Sigma is mostly about reducing spread, so these get as much attention as the mean.
- Position: percentiles and quartiles, used in box plots and to state service levels ("90% of orders ship within 2 days").
- Shape: skewness and the overall pattern seen in a histogram. See statistical distributions for the common shapes.
More on these measures is on the basic statistics page.
Common analytical (inferential) statistics
- Confidence intervals give a range that is likely to contain the true population value, such as the true mean cycle time.
- Hypothesis tests (t-tests, chi-square, ANOVA and their non-parametric versions) check whether an observed difference is larger than chance alone would explain.
- Regression estimates how an output Y changes with inputs X, which is the Y = f(X) question at the heart of the Analyze phase.
- Capability indices (Cp, Cpk) predict how much future output will fall inside specification, which only makes sense if the process is in statistical control.
Inference depends on the sample. The central limit theorem explains why sample means behave predictably, and the sample size guide shows how many data points you need for a given margin of error.
Worked example: one sample, both kinds of statistics
The numbers below are illustrative, chosen to show the method; they are not data from a real company.
- Data. A team times 10 randomly chosen orders (minutes): 12, 15, 11, 14, 13, 18, 12, 16, 14, 15.
- Describe. The sum is 140, so the mean is 140 ÷ 10 = 14.0 minutes. Sorted, the values are 11, 12, 12, 13, 14, 14, 15, 15, 16, 18, so the median is 14.0. The range is 18 − 11 = 7. The squared deviations from 14 add up to 40, so the sample variance is 40 ÷ 9 = 4.44 and the standard deviation is 2.11 minutes. These numbers are exact for these 10 orders.
- Analyze: confidence interval. The standard error is 2.11 ÷ √10 = 0.667. With 9 degrees of freedom the 95% t value is 2.262, so the interval is 14.0 ± 2.262 × 0.667 = 14.0 ± 1.51, or 12.5 to 15.5 minutes. We are 95% confident the true mean cycle time of the process is in that range.
- Analyze: hypothesis test. The target is under 15 minutes. A one-sample t-test gives t = (14.0 − 15) ÷ 0.667 = −1.50 and a one-sided p-value of about 0.08. At the usual 0.05 level, the data do not prove the process beats the target, even though this sample's average does. The team would collect more data before claiming success.
The example shows why both are needed: the descriptive mean of 14.0 looks good, but the analytical step shows the evidence is not yet strong enough.
Where each is used in DMAIC
| Phase | Descriptive | Analytical |
|---|---|---|
| Measure | Baseline mean, spread and shape of the output; histograms and run charts | Check the measurement system and baseline capability |
| Analyze | Stratify the data, Pareto the causes | Test which inputs really affect the output (tests, ANOVA, regression) |
| Improve | Compare before and after summaries | Confirm the improvement is statistically significant, often with a designed experiment |
| Control | Dashboards and visual boards | Control charts separate common and special cause variation |
The phases are covered in more detail in the DMAIC guide.
Enumerative and analytic studies (Deming)
W. Edwards Deming drew a related line between enumerative studies, which describe a fixed, existing group (for example, this month's shipped lots), and analytic studies, which aim to predict how a process will behave in the future so it can be improved. Most Six Sigma work is analytic: the goal is to change the process, not just to count last month's results. That is why control charts and process stability matter before any prediction is trusted.
Common mistakes
- Treating a sample average as the truth. A mean from 10 points can move a lot with the next 10. Report a confidence interval.
- Inferring from a biased sample. Timing only the easy orders, or only the day shift, makes any inference about the whole process wrong.
- Using the mean on skewed data. For cycle times and waiting times, report the median and percentiles as well.
- Predicting from an unstable process. Capability and forecasts assume the process is in control. Check a control chart first.
- Confusing "not significant" with "no effect". A high p-value can simply mean the sample was too small.
Frequently asked questions
What is the difference between descriptive and analytical statistics?
Descriptive statistics summarize the data you collected, such as its mean, median and standard deviation. Analytical (inferential) statistics use that sample to draw conclusions about the wider population or process, with a stated level of confidence, using tools such as confidence intervals and hypothesis tests.
Is analytical statistics the same as inferential statistics?
In Six Sigma training material the terms are used interchangeably: both mean drawing conclusions beyond the sample. Descriptive statistics are likewise sometimes called enumerative statistics.
What are examples of descriptive statistics in Six Sigma?
The mean, median, mode, range, standard deviation, variance, percentiles, and charts such as histograms, box plots, dot plots and Pareto charts used to baseline a process in the Measure phase.
Which comes first, descriptive or analytical statistics?
Descriptive statistics come first. You look at the shape, center and spread of the data and check for errors or outliers before choosing and running an analytical test.
See also: basic statistics, statistical distributions, hypothesis testing, sample size, and normality and data transformation.
Last updated: · Published by Lean Six Sigma Training Guide (sixsigmatrainingfree.com) · How this guide is written