0Quick Answer

1What Is Stratified Random Sampling?

Stratified random sampling is a probability sampling method in which the population is first divided into non overlapping subgroups called strata, and then an independent simple random sample is drawn inside each stratum. The stratum estimates are combined using weights equal to each stratum share of the population. The point of the design is not fairness or coverage; it is variance reduction.

The mechanism is simple. Total variability in a population can be split into variability between the strata and variability within them. Because you sample inside every stratum, the between stratum part never enters the sampling error at all. So the better your strata separate the population, the more variance you remove for free. Stratifying a forest by habitat type, a district by income band, or a lake by depth zone routinely halves the standard error for the same number of units.

The cost is information. You need to know which stratum every unit belongs to before you sample, and you need the stratum sizes Nₕ to compute the weights. If that information does not exist in the frame, stratification is not available to you and you fall back on simple random or systematic sampling.

Figure 1.1 An independent random sample is drawn inside every stratum, then combined using the stratum weights.

2Sampling Setup: Enter Your Strata and Design

Strata: one card per stratum

Each card is one stratum. Enter its population size Nₕ, its measurement values, and a cost per unit if you are using optimal cost allocation. Stratum names are editable.

3Results

4Interpretation of Results in Detail

How to read each number

The stratum weight Wₕ. This is Nₕ divided by N, the share of the population living in that stratum. It is the single most important number in the whole design, because every estimate is a weighted combination using it. A stratum holding 5 percent of the population contributes 5 percent to the overall mean no matter how many units you sampled in it. This is why you must never average the stratum means directly; a plain average silently assumes all strata are the same size.

The allocation nₕ. This is how many units each stratum receives out of your total n. Proportional allocation gives every stratum the same sampling fraction, which makes the design self weighting and the arithmetic simple. Equal allocation gives every stratum the same count, which is right when you want to compare strata against each other rather than estimate the population total. Neyman allocation sends units to strata that are both large and variable, which minimises the variance of the overall estimate for a fixed n. Optimal cost allocation does the same but discounts strata that are expensive to reach.

The stratum mean and standard deviation. Each stratum is estimated on its own, from its own simple random sample. The standard deviation Sₕ describes spread inside that stratum only. Good stratification produces small Sₕ values and clearly different stratum means; that pattern is exactly what generates the precision gain.

The stratified mean. This is Σ Wₕ ȳₕ, the weighted combination. It is unbiased for the population mean provided the strata are exhaustive, non overlapping, and the Nₕ values are correct. An error in a single Nₕ biases the whole estimate, which is why frame accuracy matters more here than in simple random sampling.

The stratified standard error. This comes from Σ Wₕ² (Sₕ² / nₕ)(1 − nₕ/Nₕ). Read the structure: each stratum contributes its own variance, scaled by the square of its weight, and reduced by its own finite population correction. Two consequences follow. A large stratum contributes heavily because Wₕ is squared, and a stratum you sampled almost completely contributes almost nothing because its correction approaches zero.

The design effect. This compares the stratified variance to what simple random sampling would have produced with the same total n. Below 1 means stratification worked and you gained precision. Around 1 means the strata do not actually differ and you did the extra work for nothing. Above 1 is rare and usually means the allocation was badly mismatched to the stratum variances, for instance equal allocation applied to strata of very unequal size.

The effective sample size. This is n divided by the design effect. It tells you how large a simple random sample would have to be to match your precision. If you sampled 60 units with a design effect of 0.55, your effective sample size is about 109, so the stratification bought you the equivalent of 49 extra units at no field cost.

The margin of error and confidence interval. The margin of error is the critical t value multiplied by the stratified standard error, with degrees of freedom taken as n minus the number of strata, because one mean was estimated in each stratum. Report the interval, never the point estimate alone.

The minimum units warning. A stratum with only one sampled unit has no within stratum variance estimate, so its contribution to the standard error cannot be computed and the whole interval becomes invalid. The tool enforces a floor of at least two units per stratum and tells you when the allocation had to be adjusted to respect it.

What the numbers cannot tell you. None of these statistics can tell you whether your stratification variable was the right one. If you stratified by district but the real variation runs by altitude, the design effect will sit near 1 and you will have gained nothing. Choosing the stratification variable is a subject matter decision, not a statistical one, and it is the decision that determines whether this design was worth using at all.

5How to Write Your Results in Research

Use the templates below. Each one names the strata, the allocation rule, the weights and the design effect, which is exactly what a reviewer checks in a stratified design.

Rules that make a results paragraph pass review

1. Name the stratification variable in the first sentence. Write "stratified by habitat type", not "stratified sampling was used". The variable is the design decision that matters.
2. Give a table of Nₕ, Wₕ, nₕ and Sₕ. Reviewers of stratified designs expect this table; a sentence is not a substitute.
3. State the allocation rule explicitly. "Proportional allocation" and "Neyman allocation" produce different numbers, so the reader cannot reproduce your work without knowing which you used.
4. Say the strata were exhaustive and non overlapping. One clause is enough, but it must be there, because it is the core assumption.
5. Report the weighted mean, not a plain average. Say "weighted by stratum size" out loud, because unweighted averaging is the most common error in student and consultancy reports.
6. Give the design effect or the precision gain. This is the justification for having stratified at all, and it is a genuine result.
7. State the degrees of freedom as n minus L. Using n minus 1 overstates your precision slightly and reviewers do notice.
8. Report any stratum that hit the minimum size floor. If a small stratum was topped up to 2 units, say so; the allocation then deviates from the stated rule.
9. Match decimal places to the standard error. One more decimal than the SE is the convention.
10. Archive the stratum assignment. Deposit the frame with the stratum label attached to every unit, plus the seed and the drawn IDs.

Common wording mistakes and the fix

6Formulas Used

7How to Use This Tool

1. Type your study area or project name so it appears in every template and export.
2. Pick a sample dataset preset to see how the tool behaves, or go straight to your own strata.
3. Enter the total sample size n you can afford across all strata combined.
4. Choose the allocation method: proportional for a population estimate, equal for comparing strata, Neyman for maximum precision, or optimal cost when strata differ in expense.
5. On each stratum card, type the stratum name and its population size Nₕ.
6. Paste the measurements for that stratum as comma separated numbers, or use Summary only mode and enter Sₕ.
7. Add a cost per unit on each card if you selected optimal cost allocation.
8. Or switch to the upload tab, choose a CSV or Excel file, then click the columns that should each become a cluster.
9. Click Allocate, Draw and Calculate, then read the allocation table, the stratified mean and the four plots.
10. Download the Field Pack for the field day and copy the ready made methods paragraph into your manuscript.

8Detailed Reference Tables

Table 8.1 Choosing an allocation method

Table 8.2 Worked allocation, n = 100 across three strata

Neyman sends most units to stratum B because it is both the largest and by far the most variable. That is the whole idea of optimum allocation.

Table 8.3 Expected precision gain by how different the strata are

Table 8.4 Critical t values by degrees of freedom (n − L)

9Example Results (8 Worked Cards)

10How to Collect Raw Data in the Field

10a. The 18 point field protocol for stratified random sampling

Plan

1. Choose the stratification variable before anything else. Pick the one factor you believe drives most of the variation: habitat type, income band, depth zone, irrigation status. This single choice decides whether the design gains anything.
2. Make the strata exhaustive and non overlapping. Every unit in the frame must belong to exactly one stratum, with a written rule for boundary cases.
3. Count Nₕ for every stratum and check the total equals N. A wrong stratum size biases the weighted mean directly, which no amount of sampling will correct.
4. Run a pilot of 8 to 10 units per stratum if you plan Neyman allocation, because Neyman needs an Sₕ estimate that proportional allocation does not.
5. Allocate, then check every stratum has at least 2 units and draw a 20 percent reserve list inside each stratum separately.

Figure 10.1 Each stratum receives its own independent random draw, never one draw across the whole area.

Kit

6. Carry the printed Field Pack with a separate page per stratum, each showing its own drawn list and reserve list.
7. Carry a stratum boundary map or the stratum rule in writing, so a field assistant can decide which stratum a borderline unit belongs to without guessing.
8. Carry colour tags or different flagging tape per stratum, which prevents units being recorded against the wrong stratum.
9. Carry two pencils and a waterproof sheet cover. Ink runs in rain and a lost sheet is a lost field day.

Locate

10. Finish one stratum before starting the next where travel allows, so the stratum label on every sheet is unambiguous.
11. Go to the drawn unit, not the convenient one, and never move a unit across a stratum boundary because it is easier to reach.
12. If a unit is inaccessible, replace it from the reserve list of the same stratum only. A replacement from a different stratum corrupts both weights.

Figure 10.2 A replacement must come from the reserve list of the same stratum, never from a neighbouring one.

Collect

13. Write the stratum label on every single row, not just once at the top of the sheet. Sheets get separated and rows get retyped.
14. Record a true zero as 0, never as a blank. A blank means not measured; a zero means measured and none present.
15. Record effort per stratum as well as per unit, because unequal cost between strata is what justifies optimal cost allocation next season.

Figure 10.3 A completed row with the stratum recorded twice, in the label column and in the unit ID.

Figure 10.4 Three records that must never be written the same way; in a stratified design each one changes nₕ differently.

Check

16. At the end of each stratum, count rows against the allocation. Rows plus Not found entries must equal nₕ exactly for that stratum.
17. Confirm no stratum finished with fewer than 2 measured units. If one did, either return for more or merge it with the most similar stratum and say so in the methods.
18. Photograph every sheet the same evening, keeping the sheets grouped by stratum, and enter the data within 48 hours.

Figure 10.5 Cost per unit recorded per stratum is what makes optimal cost allocation possible next time.

10b. Datasheet column specification

Zero rule: write 0 for measured and none present; leave blank only when the unit was not measured at all. Boundary rule: if a unit sits on a stratum boundary, assign it by the written rule and record that decision in the notes; never leave the stratum column blank.

10c. Filled worked datasheet

What this sheet shows:

• The stratum appears in its own column and again inside every unit ID, so a separated sheet can always be repaired.
• Row 5 is a true zero written as 0, so it still contributes to the stratum B mean.
• Row 6 is a non-response with the reason recorded, so the loss is auditable and nₕ can be corrected.
• Row 7 uses a reserve unit from stratum B, marked with an R prefix, which keeps both weights intact.
• The stratum A values cluster far above the stratum B values, which is visible evidence that the stratification is working.

10d. Blank print ready datasheet

Download the blank sheet as a .csv file, open it in Excel, Google Sheets or LibreOffice, then print one copy per stratum. The Stratum, Nₕ and nₕ columns are pre-labelled so completeness can be checked in the field, and the rows stay readable after a black and white photocopy.

11Which Sampling Method Should You Use?

Decision tree. Answer these in order and stop at the first yes.

1. Do you know a variable that splits the population into groups with clearly different means, and is it recorded in the frame? Use stratified random sampling.
2. Do you need a separate reportable estimate for a small subgroup? Use stratified sampling with disproportionate allocation, and remember to weight when combining.
3. Do you have pilot standard deviations for each stratum and want maximum precision? Use Neyman allocation.
4. Do strata differ a lot in travel time or measurement cost? Use optimal cost allocation.
5. Is the frame ordered along a gradient rather than split into groups? Use systematic sampling, or stratify along the gradient and sample within.
6. Is the frame complete but unstructured? Use simple random sampling.
7. Is there no unit list, but the population comes in natural groups you can list? Use cluster sampling.
8. Is the population hidden or hard to reach, with no frame at all? Use respondent driven or snowball sampling, and accept that it is non-probability.

12Troubleshooting and Common Sampling Errors

13Assumptions, Bias and Limitations

• Exhaustive and non overlapping strata. Every unit belongs to exactly one stratum. Overlap double counts units and gaps exclude them; both bias the weighted mean.
• Known and correct stratum sizes. The weights Wₕ come straight from Nₕ. An error in a stratum size biases the estimate directly and no sample size will correct it.
• Independent sampling within strata. Each stratum needs its own random draw. One draw across a pooled frame is simple random sampling wearing a stratified label.
• At least two units per stratum. With one unit the within stratum variance is undefined, so the standard error of the whole estimate cannot be computed.
• Neyman depends on the Sₕ inputs. Optimum allocation is optimal only for the variable whose standard deviations you used. It can be poor for a second variable measured on the same visit.
• One allocation cannot optimise two variables. If you measure several outcomes with different variance patterns, proportional allocation is the safer compromise.
• Non-response bias. If non-response differs by stratum, the achieved nₕ values drift away from the allocation and the weights no longer match the realised sample.
• Measurement error. These formulas describe sampling variability only. A miscalibrated instrument shifts every value and no confidence interval will warn you.

14Conclusion

What stratified random sampling gives you

Stratified sampling is the one design change that reliably buys precision without buying more units. Because you sample inside every stratum, the variation between the strata never enters the sampling error at all. If your strata genuinely separate the population, cutting the variance by a third to a half for the same field effort is routine, and the tool reports that gain explicitly as a design effect. The design also guarantees coverage of every subgroup, which matters when a small but important stratum would otherwise be missed by chance in a simple random draw.

What it costs you

The cost is information and bookkeeping. You need a stratum label on every unit in the frame before you sample, and you need accurate stratum sizes, because the weights come directly from them. You then have to carry those weights through every calculation, and the most common failure in practice is not a statistical one but an arithmetic one: someone averages the stratum means and forgets the weights entirely. Neyman allocation adds a further requirement, a pilot estimate of each stratum standard deviation, and it is only optimal for the variable those estimates came from.

What to check before you publish

Confirm five things: the strata were exhaustive and non overlapping with a written boundary rule, the stratum sizes Nₕ sum to the population total, every stratum received at least two sampled units, the mean was computed as a weighted sum rather than a plain average, and the confidence interval used degrees of freedom of n minus L. Then state the design effect, because that number is the entire justification for having stratified in the first place.

What to do next

If the design effect came out near 1, the stratification variable was the wrong choice and the fix is a better variable, not more units. If it came out well below 1, keep the same strata for repeat surveys so the series stays comparable, and consider moving from proportional to Neyman allocation now that you have real standard deviations from this round. If any stratum hit the minimum size floor, either merge it with its most similar neighbour or plan a deliberate disproportionate allocation for it next season.

15Test Yourself

16Frequently Asked Questions

17Cite This Tool

18Related Tools

• Simple Random Sampling Calculator for a seeded, duplicate free random draw.
• Systematic Sampling Calculator for interval k with a random start and a periodicity check.
• Sample Size Calculator using Cochran, Yamane and Krejcie-Morgan.

19Glossary of Terms

20References

1. Neyman, J. (1934). On the two different aspects of the representative method: the method of stratified sampling and the method of purposive selection. Journal of the Royal Statistical Society, 97(4), 558-625. https://doi.org/10.2307/2342192
2. Cochran, W. G. (1977). Sampling Techniques (3rd ed.), Chapters 5 and 5A: Stratified Random Sampling. Wiley. Publisher page
3. Tschuprow, A. A. (1923). On the mathematical expectation of the moments of frequency distributions in the case of correlated observations. Metron, 2, 461-493, 646-680. WorldCat record
4. Dalenius, T., & Hodges, J. L. (1959). Minimum variance stratification. Journal of the American Statistical Association, 54(285), 88-101. https://doi.org/10.1080/01621459.1959.10501501
5. Sethi, V. K. (1963). A note on optimum stratification of populations for estimating the population means. Australian Journal of Statistics, 5(1), 20-33. https://doi.org/10.1111/j.1467-842X.1963.tb00132.x
6. Lavallee, P., & Hidiroglou, M. (1988). On the stratification of skewed populations. Survey Methodology, 14(1), 33-43. Statistics Canada
7. Kish, L. (1965). Survey Sampling. Wiley. Archive record
8. Lohr, S. L. (2021). Sampling: Design and Analysis (3rd ed.). CRC Press. https://doi.org/10.1201/9780429298899
9. Thompson, S. K. (2012). Sampling (3rd ed.). Wiley. https://doi.org/10.1002/9781118162934
10. Sarndal, C. E., Swensson, B., & Wretman, J. (1992). Model Assisted Survey Sampling. Springer. https://doi.org/10.1007/978-1-4612-4378-6
11. Horvitz, D. G., & Thompson, D. J. (1952). A generalization of sampling without replacement from a finite universe. Journal of the American Statistical Association, 47(260), 663-685. https://doi.org/10.1080/01621459.1952.10483446
12. Hansen, M. H., & Hurwitz, W. N. (1943). On the theory of sampling from finite populations. Annals of Mathematical Statistics, 14(4), 333-362. https://doi.org/10.1214/aoms/1177731356
13. Levy, P. S., & Lemeshow, S. (2008). Sampling of Populations: Methods and Applications (4th ed.). Wiley. https://doi.org/10.1002/9780470374597
14. Groves, R. M., et al. (2009). Survey Methodology (2nd ed.). Wiley. Publisher page
15. Gregoire, T. G., & Valentine, H. T. (2007). Sampling Strategies for Natural Resources and the Environment. Chapman and Hall. https://doi.org/10.1201/9780203498880
16. Elzinga, C. L., Salzer, D. W., & Willoughby, J. W. (1998). Measuring and Monitoring Plant Populations. BLM Technical Reference 1730-1. BLM library
17. Krebs, C. J. (1999). Ecological Methodology (2nd ed.). Benjamin Cummings. WorldCat record
18. Rao, J. N. K., & Molina, I. (2015). Small Area Estimation (2nd ed.). Wiley. https://doi.org/10.1002/9781118735855
19. Wolter, K. M. (2007). Introduction to Variance Estimation (2nd ed.). Springer. https://doi.org/10.1007/978-0-387-35099-8
20. Marsaglia, G. (2003). Xorshift RNGs. Journal of Statistical Software, 8(14), 1-6. https://doi.org/10.18637/jss.v008.i14