Research Methods in Tourism, Hospitality and Events Management

4.5.2 Sample size and the calculation of sampling error

Where samples are selected on a random basis, the range of sampling error can be

calculated according to the laws of probability and statistical theory. For a full

discussion of the formulae to perform this calculation, the reader is referred to Moser

and Kalton (1993). The formula produces margins of error in percentages in relation

to the proportions of respondents answering in a particular way for a given sample

size, as indicated in 

Table 4.4

.

To understand 

Table 4.4

, consider that in a survey of visitors to a tourist attraction

50% were found to be local residents and 50% tourists. If the sample of a known

population was 30 people, the actual amount for the whole population could be

expected to be between 30 and 70% of this result (that is, plus or minus 19.6% of the

result found in the survey). If the sample size is increased to 3,000 people, the actual

value for the whole population (for a similar 50% result) could be expected to be

between 48% and 53% (plus or minus 1.8% of the result found in the survey). As the

proportions of people answering a question in a particular way change, then so does

the sampling error. For example, if it was found in a survey of 3,000 people that 90%

were in favour of providing an additional facility and 10% were against it, then the

real value for the survey population could be expected to be plus or minus 1.1% of

this figure.

Note: * Shows that the sampling error percentage is greater than the original

size

N.B. The percentages in this table are calculated to the 95% level. This means

that there is a 95 per cent probability that the percentages found in a survey lie

within a range equal to the percentage found, plus or minus the percentage

shown in the table. The table shows the range of sampling error for the results

of simple random surveys with samples of varying sizes.

---

A problem arises: on which question in a survey do we focus the sampling error to

inform the sample size? Clearly, a practical solution is to set the standard of accuracy

at the most crucial question or issue in the survey. If, for example, the main objective

of the survey is to test the different reactions to the tourist attraction by residence of

the respondents, then this is the best question to set sample size against. If a survey

has several objectives of equal importance, then the subject where there is most

disagreement or variation should be taken as the basis on which to inform sample

size.

This method assumes some knowledge of the population and furthermore stresses

the need for carefully conducted pilot surveys. What has not been mentioned in this

context is the influence of response rates on these calculations. Clearly, the accuracy

of sampling error calculations becomes increasingly suspect if non-response rates are

very high.