A sample is the list of people chosen to receive the survey.

In a stratified sample, names are grouped by segment before sampling and the same number of names is chosen from each group. This ensures that the number of responses from each segment of interest yields an acceptable sampling error.

Sampling error is the ± percentage you see associated with surveys, such as “Results are subject to a maximum sampling error of ±5% at the 95% confidence level.” This means the chances are 95 in 100 that the results are within 5 percentage points (higher or lower) of the true percentage for the entire population.

To cut sampling error in half, you need four times the responses at about four times the price! Stratifying your sample balances sampling error and your budget.

Stratifying a Sample

Stratifying a sample enables you to easily breakout survey results by different segments of your population.

Aggregated results allow you to make claims such as, “20% of respondents work for manufacturing firms.” It may be helpful to drill deeper and determine the characteristics of the manufacturing firms. You might discover that “of the 20% who work for manufacturing firms, 90% work for companies with more than 100 employees.” Then you could focus on this group's answers to other questions.

Unfortunately, if too few respondents work for a manufacturing firm, the sampling error will be so large that analysis on the segment will be impossible. Stratifying the sample can help. When a sample is stratified, various segments of interest are identified prior to sample selection and a sufficient quantity of names is selected from each.

Let’s look at the breakdown of a population by type of company:

Manufacturing: 10% - Distribution: 30% - Retail: 60%

Without stratifying the sample, you would expect to receive the same proportion of responses as the population.

Here is a scenario for a study collecting 600 total responses:

Segment	Population	Proportion Sample	Responses	Sampling Error
Manufacturing	10%	10%	60 (10% of 600)	+/- 13%
Distribution	30%	30%	180 (30% of 600)	+/- 7%
Retail	60%	60%	360 (60% of 600)	+/- 5%
Total	100%	100%	600	+/- 4%

Considering the large sampling error figure, the data shouldn't be used describe the typical manufacturing respondent. You could increase the size of the total sample by a factor of three (the number of manufacturing responses should increase threefold), but your survey now costs about three times as much.

A more attractive option is to stratify the sample—grouping the names by company type before sampling, and then selecting a specific number of names from each segment (rather than in proportion to actual population totals).

This is one possible scenario of stratifying a sample and its impact:

Segment	Population	Stratified Sample	Responses	Sampling Error
Manufacturing	10%	33%	200 (33% of 600)	+/- 7%
Distribution	30%	33%	200 (33% of 600)	+/- 7%
Retail	60%	33%	200 (33% of 600)	+/- 7%
Total	100%	100%	600	+/- 4%

Collecting an equal number of responses from each segment provides the same sampling error figure for each. The small sacrifice in precision for retail (±5% vs. ±7%) is more than offset by increased precision for manufacturing.

Now you can look at segment-level results with confidence. In the report, “total” results are weighted to true population proportions so that the categories with fewer members are not over-represented in the totals.

Stratifying your sample is an affordable way to examine the characteristics of specific segments while still getting data that accurately reflects the overall characteristics of your population.