Program Evaluation

Tipsheet #60

How to Determine a Sample Size

I want to survey a large group of people. What size should my sample be? Twenty percent? Thirty percent? AVOID
There is no set percentage that is accurate for every population. What matters is the actual number or size of the sample, not the percentage of the population. Consider a coin toss: the first few times you flip the coin, the average result may be skewed wildly in one direction (say you got 5 tails in a row), but the more times you flip the coin, the more likely that the average result will be an even split between heads and tails. So, if you surveyed 20% of a group of 300 program participants to produce a sample of 60 people, you would under represent the population, since there is a fairly large chance in a small group that the respondents you choose will vary from the whole population. On the other hand, 20% of 30,000 county residents (a sample of 6,000) would be a wastefully large sample, and not significantly more accurate than a sample of 400.

USE

There are 5 steps in deciding a sample size. If you are familiar with them, select a sample size using the tables in Appendices 1 and 2 at the end of the Tipsheet. To use a formula, move to Appendix 3. If you wish to review the five steps before selecting a sample size however, read on.

Steps in Selecting a Sample-Size

An appropriate sample size is based on a number of accuracy factors that you must consider. Together they comprise a five step process: 1. 2. 3. 4. 5. Determine Goals Determine desired Precision of results Determine Confidence level Estimate the degree of Variability Estimate the Response Rate

Step One: Determine Goals First, know the size of the population with which youre dealing. If your population is small (200
people or less), it may be preferable to do a census of everyone in the population, rather than a sample. For a marginally higher cost than a 134-person sample, you can survey the entire population and gain a 0% sampling error. However, if the population from which you want to gather information is larger, it makes sense to do a sample. Second, decide the methods and design of the sample youre going to draw and the specific attributes or concepts youre trying to measure. Third, know what kind of resources you have available, as they could be a limitation on other steps below such as your level of precision. Once you have this information in-hand, youre ready to go on to the next step.

Step Two: Determine the Desired Precision of Results

The level of precision is the closeness with which the sample predicts where the true values in the population lie. The difference between the sample and the real population is called the sampling error. If the sampling error is 3%, this means we add or subtract 3 percentage points from the value in the survey to find out the actual value in the population. For example, if the value in a survey says that 65% of farmers use a particular pesticide, and the sampling error is 3%, we know that in the real-world population, between 62% and 68% are likely to use this pesticide. This range is also commonly referred to as the margin of error. The level of precision you accept depends on balancing accuracy and resources. High levels of precision require larger sample sizes and higher costs to achieve those samples, but high margins of error can leave you with results that arent a whole lot more meaningful than human estimation. The tables in Appendices 1 and 2 at the end of the Tipsheet provide sample sizes for precision levels of 5% and 3% respectively.

Step Three: Determine the Confidence Level

The confidence level involves the risk youre willing to accept that your sample is within the average or bell curve of the population. A confidence level of 90% means that, were the population sampled 100 times in the same manner, 90 of these samples would have the true population value within the range of precision specified earlier, and 10 would be unrepresentative samples. Higher confidence levels require larger sample sizes. The tables at the end of this Tipsheet assume a 95% confidence level. This level is standard for most social-science applications, though higher levels can be used. If the confidence level that is chosen is too low, results will be statistically insignificant.

Step Four: Estimate the Degree of Variability

Variability is the degree to which the attributes or concepts being measured in the questions are distributed throughout the population. A heterogeneous population, divided more or less 50%-50% on an attribute or a concept, will be harder to measure precisely than a homogeneous population, divided say 80%-20%. Therefore, the higher the degree of variability you expect the distribution of a concept to be in your target audience, the larger the sample size must be to obtain the same level of precision. To come up with an estimate of variability, simply take a reasonable guess of the size of the smaller attribute or concept youre trying to measure, rounding up if necessary. If you estimate that 25% of the population in your county farms organically and 75% does not, then your variability would be .25 (which rounds up to 30% on the table provided at the end of this Tipsheet). If variability is too difficult to estimate, it is best to use the conservative figure of 50%. Note: when the population is extremely heterogeneous (i.e., greater than 90-10), a larger sample may be needed for an accurate result, because the population with the minority attribute is so low. At this point, using the level of precision and estimate of variability youve selected, you can use either the table or the equation provided at the bottom of this Tipsheet to determine the base sample size for your project.

Step Five: Estimate the Response Rate

The base sample size is the number of responses you must get back when you conduct your survey. However, since not everyone will respond, you will need to increase your sample size, and perhaps the number of contacts you attempt to account for these non-responses. To estimate response rate that you are likely to get, you should take into consideration the method of your survey and the population

involved. Direct contact and multiple contacts increase response, as does a population which is interested in the issues, involved, or connected to the institution doing the surveying, or, limited or specialized in character. You can also look at the rates of response that may have occurred in similar, previous surveys. When youve come up with an estimate of the percentage you expect to respond, then divide the base sample size by the percentage of response. For example, if you estimated a response rate of 70% and had a base sample size of 220, then your final sample size would be 315 (220/0.7). Once you have this, youre ready to begin your sampling! One final note about response rates: the past thirty years of research have demonstrated that the characteristics of non-respondents may differ significantly from those of respondents. Follow-up samples may need to be taken of the non-respondent population to determine what differences, if any, may exist.

Appendix 1 Example: 5% Error and Qualification. Appendix 2 Example: 3% Error and Qualification. Appendix 3 Example: An Equation for Determining Final Sample Size.

References:
Blalock, Hubert M. (1972). Social Statistics. New York: McGraw-Hill Book Company. Israel, Glen D. 1992. Determining Sample Size. Program Evaluation and Organizational Development, IFAS, University of Florida. PEOD-6. National Science Foundation, Research and Development in Industry: 1992, NSF 95-324. Arlington, VA. Smith, M.F. 1983. Sampling Considerations in evaluating Cooperative Extension Programs. Cooperative Extension Service, IFAS, University of Florida. DRAFT. Taylor-Powell, Ellen. May 1998. Sampling. Program Development and Evaluation, University of Wisconsin Extension. G3658-3. Sudman, Seymour (1976). Applied Sampling. New York: Academic Press. Warmbrod, J. Robert (1965). The Sampling Problem in Research Design. Agriculture Education Magazine. pp 106-107, 114-115. Yamane, Taro (1973). Statistics: an introductory analysis. New York: Harper & Row.
Jeff Watson, Research Assistant, Cooperative Extension & Outreach The reference citation for this Tipsheet is: Watson, Jeff (2001). How to Determine a Sample Size: Tipsheet #60, University Park, PA: Penn State Cooperative Extension. Available at: http://www.extension.psu.edu/evaluation/pdf/TS60.pdf This Web site is copyrighted by The Pennsylvania State University. The information may be used for educational purposes but not sold for profit.

Appendix 1: Tablesa for Finding a Base Sample Sizeb +/- 5% Margin of Error c Sample Size
Population 100 e 125 150 175 200 225 250 275 300 325 350 375 400 425 450 500 600 700 800 900 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 10,000 15,000 20,000 25,000 50,000 100,000 Variability 50% 40% 81 79 96 93 110 107 122 119 134 130 144 140 154 149 163 158 172 165 180 173 187 180 194 186 201 192 207 197 212 203 222 212 240 228 255 242 267 252 277 262 286 269 333 311 353 328 364 338 370 343 375 347 378 350 381 353 383 354 385 356 390 360 392 362 394 363 397 366 398 367 30% 63 72 80 87 93 98 102 106 109 113 115 118 120 122 124 128 134 138 142 144 147 158 163 165 166 167 168 168 169 169 170 171 171 172 172 20% 50 56 60 64 67 70 72 74 76 77 79 80 81 82 83 84 87 88 90 91 92 96 98 99 99 100 100 100 100 100 101 101 101 101 101 10% 37 40 42 44 45 46 47 48 49 50 50 51 51 51 52 52 53 54 54 55 55 57 57 58 58 58 58 58 58 58 58 58 58 58 58
d

Qualifications a) This table assumes a 95% confidence level, identifying a risk of 1 in 20 that actual error is larger than the margin of error (greater than 5%). b) Base sample size should be increased to take into consideration potential non-response. c) A five percent margin of error indicates willingness to accept an estimate within +/- 5 of the given value. d) When the estimated population with the smaller attribute or concept is less than 10 percent, the sample may need to be increased. e) The assumption of normal population is poor for 5% precision levels when the population is 100 or less. The entire population should be sampled, or a lesser precision accepted.

Appendix 2: Tablesa for Finding a Base Sample Sizeb +/- 3% Margin of Error c Sample Size
Population 2,000 e 3,000 4,000 5,000 6,000 7,000 8,000 9,000 10,000 15,000 20,000 25,000 50,000 100,000 Variability 50% 40% 714 677 811 764 870 816 909 850 938 875 959 892 976 908 989 920 1000 929 1034 959 1053 975 1064 984 1087 1004 1099 1014 30% 619 690 732 760 780 795 806 815 823 846 858 865 881 888 20% 509 556 583 601 613 622 629 635 639 653 660 665 674 678 10% 322 341 350 357 361 364 367 368 370 375 377 378 381 383
d

Qualifications a) This table assumes a 95% confidence level, identifying a risk of 1 in 20 that actual error is larger than the margin of error (greater than 3%). b) Base sample size should be increased to take into consideration potential non-response. c) A three percent margin of error indicates willingness to accept an estimate within +/- 3 of the given value. d) When the estimated population with the smaller attribute or concept is less than 10 percent, the sample may need to be increased. e) The assumption of normal population is poor for 3% precision levels when the population is 2,000 or less. The entire population should be sampled, or a lesser precision accepted.

Appendix 3: An Equation for Determining Final Sample Size P[1 P ] A 2 P[1 P ] 2 + N n=Z R
Where: n = sample size required N = number of people in the population P = estimated variance in population, as a decimal: (0.5 for 50-50, 0.3 for 70-30) A = Precision desired, expressed as a decimal (i.e., 0.03, 0.05, 0.1 for 3%, 5%, 10%) Z = Based on confidence level: 1.96 for 95% confidence, 1.6449 for 90% and 2.5758 for 99% R = Estimated Response rate, as a decimal