The issue of standardising data collection methods in panel research is made the more difficult and at the same time the more important by the fact that there are very few agreed upon principles in panel design. Therefore, if the current interest in making panel data comparable will result in the emergence of some basic guidelines for panel design, this would be beneficial to the whole field of panel research. In this paper, only a small, albeit important aspect of panel design will be discussed ~ namely that of panel composition.
The election polls as now operated have few prospects to offer. This is also true if we confine ourselves to the causes of error brought about by insufficiencies in the questionnaire and leave out of consideration the purely technical aspect of sampling, i.e . the margins of error which are themselves of uncertain validity. It all hinges upon a questionnaire that has been tested as to its predictive effectiveness . So long as that is not done, election polls will be nothing but a gamble instead of a serious calculation of probabilities . An important hindrance to putting efficacy research into practice is undoubtedly that the customers for election forecasts, that is to say the news media, it must be feared, have as yet. Little comprehension of them and will presumably not be able to afford much towards financing it. But so long as market research organisations continue to regard this sort of research as an attractive form of free publicity, and therefore offer it at low prices, there is unfortunately little chance of an improvement in the climate in this respect.
This paper has argued that the sample size necessary in a survey does not primarily depend upon the size of the population from which the sample is to be drawn, but from other factors concerning the basic characteristics of the population and the quality of the information required from the sample. Hence the idea of a constant percentage sample size is a myth; there never has been, or will be, a sampling plan that requires a constant percentage of the population to be sampled, valid for all population sizes. To determine the correct sample size, for defined precision and confidence, some knowledge is required of P for attribute sampling or S for measured variates. This knowledge is not always available and thus, it may sometimes be argued, the procedures outlined earlier are vitiated. But, whilst it is often true that P or S are not precisely known, an approximate value is commonly available and this may well be good enough. Furthermore, a small error in P or S may not vitally affect the value of N obtained. If a big error is found, this would become clear as the sampling progressed and, should circumstances warrant it, a re-calculation of N can made and the remaining balance of the freshly estimated sample size N obtained on a statistical basis. Finally, this paper has only discussed straightforward single sample plans. Whilst the basic thesis of the paper holds whatever the kind of sampling plan envisaged, e.g. double sampling, or stratified sampling, the determination of the necessary sample size becomes more involved and the reader must be referred to the more specialised standard works on sampling, whether for the general background, the mathematical theory, or the survey design and analysis necessary for rigorous sampling.
The two approaches had been introduced as alternatives. As such, the group agreed with Ehrenberg & Goodhardt's approach. They did, so on purely practical grounds . The crucial drawback seen in Hamre's approach was that the number of constants or coefficients to be fitted from the data kept increasing. The group felt that this would lead to instability under sampling variation.