by Steve Langton, Defra Environmental Observatory, 1-2 Peasholme Green, York YO1 7PX, UK.
Surveys are widely used in many areas of modern life. Political opinion polls and the myriad of phone and postal surveys aimed at the general public spring instantly to mind. There are also vast numbers of more specialized surveys aimed at producing key facts for business, government, medical researchers and others. In addition, many scientific studies involve random sampling and may require the use of survey analysis methods.
The analysis of surveys is, in many cases, a fairly simple exercise compared to many other statistical analyses. Unfortunately that simplicity often tempts analysts to rely on unsuitable software, such as simple spreadsheet programs. Whilst these often give correct point estimates, they seldom produce valid standard errors and do not provide a means of identifying outlying or influential observations. The aim of this Guide is to show how the correct analysis can easily be achieved using GenStat’s facilities for survey analysis.
GenStat can be used in two ways; the simplest, particularly for new users, is to use the menu system, and this Guide will show you how to perform all the analyses using menus. The second way is to use GenStat’s own programming language, and this can be an efficient approach for many surveys since it allows the automation of repetitive tasks. The use of programming is not described in the main text, but a separate chapter introduces the principles and some key commands, whilst an Appendix gives the commands to generate all the analyses described in the main text. Those keen to learn to program in GenStat may prefer to read the programming chapter first and then refer to the Appendix whilst working through the earlier chapters.
The first stage in any survey is the design phase, but in this book we will concentrate on survey analysis, only briefly considering design issues. This should not be taken to imply that the design of a survey is not crucially important, but instead is a pragmatic decision based on the knowledge that many GenStat users will have to analyse surveys which they have not had the opportunity to design.