The 5th and 95th percentiles describe the lower and upper tails of MDF test results, helping you see variation that an average can hide. For purchasing, match the reported property, sampling unit and calculation method to the acceptance rule in your specification. Our factory team helps connect your MDF selection and required test records with the component you are ordering; a percentile alone is not a finished-part load rating.
Have a report or buyer specification to interpret? Start with the property and the part you are making. We help identify the relevant board, sample plan and reporting requirement for your order.
Plain MDF sheets Custom-size MDF parts
What the lower and upper percentiles tell you
For a continuous distribution, the 5th percentile has 5% of values below it and 95% above it; the 95th percentile reverses those proportions. A lower-tail value can be relevant when a high result is favourable, as with some strength properties. An upper-tail value can be relevant when the requirement limits a high result, as with a specified swelling measure. The applicable product and assessment rules decide which statistic is required.
NIST's percentile guidance also explains that sample percentile calculations can use different interpolation conventions. Therefore a spreadsheet result needs its method recorded, particularly for small datasets where the tail estimate is sensitive to that choice.
Distinguish percentiles, confidence and tolerance limits
| Statement | What it describes | Question to ask |
|---|---|---|
| Observed minimum | The lowest result in this sample | How were the samples selected? |
| Estimated 5th percentile | An estimate of a lower distribution quantile | Which estimator and population model were used? |
| Confidence interval for the mean | Uncertainty in estimating the population mean | Why is the mean the quantity of interest here? |
| Statistical tolerance limit | A bound designed for population coverage at a stated confidence | What coverage, confidence and sample size were specified? |
A result labelled only 95% is incomplete. Ask whether the percentage refers to a percentile position, confidence, coverage or another reported quantity. These meanings should appear explicitly in the laboratory or statistical summary, not be inferred from a column heading.
Calculate the tails of a normal distribution
When a normal-distribution model is appropriate, the population percentiles can be calculated as q05 = μ − 1.64485σ and q95 = μ + 1.64485σ, where μ is the population mean and σ is its standard deviation. NIST's normal-distribution reference explains this relationship.
| Input or result | Meaning for the comparison |
|---|---|
| Mean, μ | The centre of the population described by the model. |
| Standard deviation, σ | Spread around that centre; larger spread moves both tails farther from the mean. |
| 5th percentile, q05 | The lower tail relevant to properties where a minimum matters. |
| 95th percentile, q95 | The upper tail relevant to properties where a maximum matters. |
| Data and method | The underlying measurements, sample size, model and applicable test or product standard. |
Equal averages can accompany different lower-tail performance. Compare variation and the same reported statistic for each product, then use the specified sampling and assessment method for the procurement decision.
Understand where the 1.645 factor belongs
In practice, the population mean and standard deviation are usually unknown. Replacing them with sample estimates introduces uncertainty. An estimated normal percentile and a confidence-qualified statistical tolerance limit answer different questions.
NIST's normal tolerance-interval guidance describes bounds of the form sample mean plus or minus a factor times sample standard deviation. That factor depends on the specified coverage, confidence, sample size and method. A one-sided bound and a two-sided interval also require different treatment. Statistical tolerance limits are separate from engineering acceptance limits.
For MDF conformity calculations, use the complete applicable standard and the laboratory's stated procedure. Carry its factor, grouping rule and any minimum variability provision into the calculation, with enough detail for your receiving team to understand the result. An isolated formula or incomplete table does not supply those assessment rules.
Connect test pieces to boards and lots
Ten test pieces cut from one board provide information about that sampled board; they are not automatically ten independently sampled boards from the shipment. A dataset should preserve the relationship between each specimen, its parent board and the production lot. Otherwise the apparent sample size may exaggerate the evidence about between-board variation.
EN 326-1 addresses sampling, cutting test pieces and expressing results for wood-based panels. Its stated scope reinforces the need to define how specimens become reported results. Use the relevant full procedure and agreed sampling plan rather than inventing an acceptance count.
Keep different grades, thickness groups, conditioning procedures and test methods identifiable. Keeping those groups separate makes the result useful for the actual grade and thickness being ordered. If a combined analysis is required, document its justification and retain the separate underlying results.
Check the data behind the result
Review specimen IDs, units, transcription, conditioning and any invalid-test records. Investigate unusual results; do not discard a low strength result merely because it worsens the summary. An exclusion needs a technical reason, a record and the treatment prescribed by the applicable procedure.
Look for grouping or trends that challenge the assumed population model. A distribution assembled from two different products may not be usefully represented by one normal curve. The sampling plan remains important even when a result is reported to several decimal places. Where the evidence is insufficient, request a suitable analysis or additional sampling instead of attaching certainty to the existing number.
Keep the calculation and sampling details together
- Product, thickness, lot identity, property and units.
- Sampling plan, number of sampled boards and number of specimens per board.
- Test method, conditioning and the rule for combining specimen results.
- Individual results, treatment of invalid tests and the reported summary statistic.
- Percentile estimator or statistical model, with coverage and confidence if a tolerance limit is reported.
- Applicable specification, edition and the exact basis for the conformity decision.
The official record for GB/T 11718-2021 identifies the published MDF standard. An edition identifier matters: an incomplete excerpt with a different identifier is not a substitute for the required assessment text.
Use the statistic in your MDF specification
Start with the required property and acceptance basis, then decide what evidence is needed. For a structural application, assess the relevant design information alongside the specified material statistic. Our MDF properties guide explains what each physical or mechanical metric measures.
Use our incoming-inspection guide to connect reports to shipment identity. When you send your MDF enquiry, include the required grade, thickness, destination and reporting method so we can discuss evidence in the same terms as your specification.