Several experiments yield data on an ordered categorical scale such as very good, good, moderate, poor and very poor. One of the critical issue which frequently arises when setting up a questionnaire with Likert scales, concerns the appropriate number of categories m for a given sample size n. Increasing the number of categories enhances the eciency of the estimators supporting the choice of a large value for m. However, if the sample size is small, when the number of options increases one or more categories may have no observations. Under the proportional odds assumption with probit link, the suitable number of options is investigated through a simulation experiment concerning several combinations of sample size and number of categories. The results make clear that to compensate limited sample information due to a small n, m needs to be large.
The suitable sample size for ordinal categorical data when setting the number of categories
Monti Anna Clara;
2019-01-01
Abstract
Several experiments yield data on an ordered categorical scale such as very good, good, moderate, poor and very poor. One of the critical issue which frequently arises when setting up a questionnaire with Likert scales, concerns the appropriate number of categories m for a given sample size n. Increasing the number of categories enhances the eciency of the estimators supporting the choice of a large value for m. However, if the sample size is small, when the number of options increases one or more categories may have no observations. Under the proportional odds assumption with probit link, the suitable number of options is investigated through a simulation experiment concerning several combinations of sample size and number of categories. The results make clear that to compensate limited sample information due to a small n, m needs to be large.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
