Run this study#
Your question: what do these results look like, and how much do they vary?
Bring: the numeric results to describe, one per row, the analyte and its unit. Keep coded specimen IDs when the same specimen appears more than once. There is no fixed size. Bring every result that fits your selection rule. For example, 30 specimens measured once give 30 rows; two groups of 30 give 60 rows, summarized separately.
- Decide what belongs in the summary. Write a selection rule, such as all eligible specimens from a stated period. Put different populations in separate levels. Use real specimen IDs so repeats show.
- Set any limit first. A description needs no limit. If your laboratory has one, add it on Set up under Acceptance limits before you calculate: Enter your own limit offers SD, CV and Mean within a range, and More statistics adds median, minimum and maximum. See Acceptance limits.
- Import the results. A Result column is enough when each row is a different specimen; the app generates its IDs. Include Specimen / material ID for repeated specimens and Level for separate groups. Use Blank template (CSV) or descriptive.csv as the layout. On Data, choose Import a file, answer any question the dialog asks, then choose Import into this study. Check the unit, decimal separator and number of included rows.
- Review the summary. Choose Calculate results. Results opens. Read the mean, median, SD, range and plots, and look into unusual values. Excluding a value needs a reason.
- Report. On Report, choose Download PDF or Print.
What you get: a description of the results you entered. The calculation minimum is one result for a summary, two for an SD and two different specimen IDs for a limit. For precision, use repeatability or multi-day precision.
Purpose#
Summarises a set of results level by level and checks any limit you set.
When to use it#
Use it for a quick look at a data set or for results that fit no other study.
Use a different study for:
- replicates of one control in one run: Repeatability.
- replicates over several days and runs: Multi-day precision.
- comparison with a known value: Bias against an assigned value.
Study setup#
- Choose New study. Enter Analyte and Unit, and select Descriptive summary under Summaries.
- Choose Create study, or Create and import data if you already have the results.
The experimental unit is the specimen. The statistics use every included row, but a limit needs at least 2 different specimen IDs. Repeated IDs turn off the mean interval, because repeat measurements of one specimen aren't independent.
Entering data#
On Data, choose Import a file or Add row. Blank template (CSV) downloads the columns this study needs.
| Column | Required? | What to enter |
|---|---|---|
| Specimen / material ID | Manual rows and repeated specimens | A coded ID, for example 00001, with no patient identifiers |
| Result | Yes | The measured value. Negative values need Permit negative results |
| Level | No | A group name, for example 1 or High. A file without a Level column puts every row in level 1 |
| Day, Run, Replicate | No | For your records |
Correct a censored value such as <5 to its exact value if you have one, or exclude the row with a reason. See Missing and censored values.
Calculation stops on a marked row (see Validation messages), a blank analyte or unit, or when every row is excluded. See Calculating results.
Statistics#
For each level:
| Estimate | How it is calculated |
|---|---|
| Mean | Arithmetic mean |
| Median, 25th / 75th percentile | Linear interpolation (Hyndman–Fan type 7) |
| Sample SD, variance | Centred sample SD (n − 1 denominator). Variance = SD² |
| CV | 100 × SD / mean. Needs a ratio scale, at least 2 results, a positive mean and no negative results |
| Minimum, Maximum | Smallest and largest included result |
| Mean CI | mean ± t(1 − α/2, n − 1) × SD / √n |
The plots include every point. The Q–Q plot and the box plot need at least 3 results that are not all identical. Q–Q points use Blom positions (i − 3/8)/(n + 1/4). Box whiskers reach the most extreme values within 1.5 × IQR of the quartiles. The histogram uses ceil(√n) equal-width bins, at most 40.
Acceptance limits for this study#
The unrounded estimate is compared with the limit, and a result exactly on the limit meets it. A CV limit is in %; the others are in the study unit, and a limit in another unit is Undecided. A level with fewer than 2 different specimen IDs, or an estimate that can't be calculated, is also Undecided. On an interval scale, where CV and percentage limits are not calculated, use an absolute limit in the study unit.
One failing limit fails the whole study. See Acceptance limits for the choices and how outcomes are decided.
Worked example#
descriptive.csv holds five results (2, 4, 6, 8, 10) from specimen IDs 00001 to 00005, all in level 1.
Settings:
- analyte
Example analyte, unitmmol/L, 3 display decimals - 95% confidence
- limits CV at most
60% and Mean from5to7mmol/L
What the saved result shows#
| Estimate | Displayed value |
|---|---|
| Mean | 6.0 mmol/L |
| Median | 6.0 mmol/L |
| Sample SD | 3.16 mmol/L |
| CV | 52.7% |
| Sample variance | 10.0 mmol/L² |
| Minimum / Maximum | 2.0 / 10.0 mmol/L |
| 25th / 75th percentile | 4.0 / 8.0 mmol/L |
| 95% CI for mean | 2.1–9.9 mmol/L |
| Acceptance limit | Observed | Limit | Outcome |
|---|---|---|---|
| CV | 52.7% | ≤ 60% | Met |
| Mean | 6.0 mmol/L | 5–7 mmol/L | Met |
The study status is Criteria met. Without limits, the report's outcome reads Descriptive – no criteria. The box plot shows quartiles 4, 6 and 8 with whiskers at 2 and 10.
Reading the results#
- Results and acceptance: one row for each limit, then the other estimates (mean, median, SD, CV, minimum, maximum). An estimate that has a limit appears only in the limit table.
- Measurements: results in entry order with the mean as a dashed line. Look for drift or a sudden shift. Histogram switches the plot to bars; Run order switches back. Select a point to find its row.
- Distribution diagnostics: a Q–Q plot and a box plot with every point.
- Additional statistics and calculation history: every statistic, including variance, quartiles and the mean CI, and a picker for earlier calculations.
The confidence interval shows where the true mean is likely to lie. Five results give a wide one, 2.1–9.9 mmol/L around 6.0.
When a limit fails#
- Recheck data entry for a misplaced decimal, a wrongly imported column or the wrong decimal separator. Check whether specimen IDs repeat and whether the scale is set correctly.
- If the spread is real, that is the finding. Check the Measurements plot for drift or an odd result.
Common mistakes#
- CI reads "Not estimated": the notes give the reason, such as repeated IDs, one result or identical values.
- One ID on every row: replicate measurements of one material belong in Repeatability.
- CV not estimated: check Measurement scale under Calculation settings, and look for negative results.
Statistical detail#
Sources: NIST/SEMATECH e-Handbook §1.3.5.6 and §1.3.5.2, NumPy quantile documentation, Blom (1958), and Hyndman and Fan (1996). See Methods and sources.