Morris H. Hansen

The Problem of Non-Response in Sample Surveys

Abstract The mail questionnaire is used in a number of surveys because of the economies involved. The principal objection to this method of collecting factual information is that it generally involves a large non-response rate, and an unknown bias is involved in any assumption that those responding are representative of the combined total of respondents and non-respondents. Personal interviews generally elicit a substantially complete response, but the cost per schedule is, of course, considerably higher than it would be for the mail questionnaire method. The purpose of this paper is to indicate a technique which combines the advantages of both procedures. The principle followed is to mail schedules in excess of the number expected to be returned, and to follow up by enumerating a sample of those that do not respond to the mail canvass. Under reasonable assumptions as to the relative costs of the two methods of canvass, an allocation of the sample can be made to mail and field canvasses. An illustration i...

An Evaluation of Model-Dependent and Probability-Sampling Inferences in Sample Surveys

Abstract In this paper we are concerned with inferences from a sample survey to a finite population. We contrast inferences that are dependent on an assumed model with inferences based on the randomization induced by the sample selection plan. Randomization consistency for finite population estimators is defined and adopted as a requirement of probability sampling. A numerical example is examined to illustrate the dangers in the use of model-dependent estimators even when the model is apparently consonant with the sample data. The paper concludes with a summary of principles that we believe should guide the practitioner of sample surveys of finite populations.

THE ESTIMATION AND INTERPRETATION OF GROSS DIFFERENCES AND THE SIMPLE RESPONSE VARIANCE

This chapter presents the estimation and interpretation of gross differences and the simple response variance. The development of theory and methods for the measurement of accuracy of survey results serves three needs. One need is to provide the planner and designer of surveys with inputs for improving the work. A second is to control the survey process in order to provide some assurance that field workers, respondents, and data processors are operating in accordance with the specifications provided them. A third need is to evaluate the results of the survey so as to provide the users with information to guide them in the decision processes in which the results of the survey are employed. While measurement of accuracy should be an important guide to the users of data, the education of users in the proper interpretation of accuracy measurements has not been sufficiently effective. Continued effort in this type of educational work is sorely needed.

The Redesign of the Census Current Population Survey

Abstract In February 1954 a redesign of the Current Population Survey was introduced that provided for a more efficient system of field organization and supervision as well as some advances in methods. The sample is now spread over 230 areas instead of 68 areas with the same number of households as heretofore. A composite estimation procedure has been introduced which reduces the sampling variability for most estimates. Also, there has been a considerable reduction in the variance of variance estimates made from the sample. A statistical quality control program has been introduced to help insure results of consistently acceptable quality. Problems arose in the process of shifting from one design to the other that resulted in some significant differences between the new and old samples for a few of the estimates, especially unemployment. Apparently response errors were the principal source of difficulty, and it was possible to take steps to bring the results within sampling error range. Work on the measure...

Insuring confidentiality of individual records in data storage and retrieval for statistical purposes

Much has been written about the question of privacy and the need for the protection of confidentiality of individual records in data storage and retrieval systems. The ability to insure confidentiality is a prime tool in the protection of privacy. The goal of this paper is to summarize from the point of view of a statistician some of the aspects and principles of confidentiality and some of the implications of these principles for computer-based storage and retrieval systems for statistical purposes. The remarks will have special relevance to open retrieval systems, that is, retrieval systems in which customers for information retrieval are the general public, or perhaps specified agencies or groups or individuals, and these customers can retrieve any desired statistics from the confidential records in the files subject to a review to insure that the output conforms to prescribed rules designed to avoid disclosure of individual information. These rules may be concerned with the minimum number of cases on which an individual statistics or frequency count is based or with other aspects, as is discussed later. The access to the data may be restricted to certain authorized types of data through control passwords or keys.

Problems and Methods of the Sample Survey of Business

Abstract In business sampling, as in the general sampling problem, the over-all test that we apply to a sample design is that it shall yield the desired information with the reliability required at a minimum cost; or, conversely, that at a given cost it shall yield the estimates desired with the maximum reliability possible. A second criterion which we impose on the sample design is that the reliability of the sample results should be measurable. This requirement eliminates from consideration many superficially attractive sampling schemes which, while generally economical in both time and money, may lead to faulty conclusions. A particularly important characteristic of the distribution of sales of business establishments is that the distribution is highly skewed, with relatively few large establishments accounting for a substantial proportion of total sales. The sample design takes advantage of this fact and in so doing yields a sample of greater efficiency than would result from a sample of equal size if...

The Development of Sample Surveys of Finite Populations

This paper discusses selected topics in the theory and methods of sample surveys which have as their objectives to provide inferences about some characteristics of a finite population (as distinguished from inferences about a causal system which may have produced certain characteristics of the finite population). The paper gives an account of some early developments, considers criteria to guide the choice of sample design, and considers various aspects of probability-sampling designs, including nonsampling error and total survey design. Finally, the paper discusses the role of models in the theory and practice of sample surveys for making inferences about a finite population, and lists some areas for research and development.

Control of Errors in Surveys

In any process of data collection there are potential sources of error at every stage in the survey procedure. Errors may arise in defining the problem, in defining the universe to be studied, and in defining the concepts or establishing the measurement procedures (such as the question wording). They may originate in the sampling, i.e., in the specification of the units to be included and in the coverage of these units. In surveys conducted by field interview, they may arise from the complicated structure of the interview situation, in some measure stemming from the understanding, interest, motivation, knowledge, and skill of the interviewer and the respondent. Procedures for handling the data, such as coding and editing and tabulation, may also lead to errors in surveys. It is the purpose of good survey design to control the errors arising from these sources to an economic level. This level is reached when an increase in expenditures will not produce a worthwhile decrease in the risk of making wrong decisions from the survey results. Although we have not found a way to determine objectively this optimum level of control in connection with general purpose surveys in which a number of different statistics are produced and used for many different purposes, we do give the problems continuing attention. The problem of control is one of particular concern because errors arising from the various possible sources, especially those arising in the field collection of the data, sometimes are much larger than is commonly recognized. Often they can be controlled satisfactorily, if at all, only by explicit steps taken for such control. It is easy, on the other hand, after one becomes acquainted with the frequency and magnitude of individual errors, to be unduly pessimistic about the value of census or survey results. With reasonable procedures for control, and with the tendency for some of the types of errors to be more or less compensating, the net effects may be small enough so that the statistics will serve adequately many different purposes. In this setting, it becomes important to establish procedures for evaluating and for controlling the errors that may arise from various sources. We shall confine our discussions primarily to the errors arising

Regression Estimates in Federal Welfare Quality Control Programs

Abstract States that administer federal family welfare programs review samples of the beneficiaries to guide corrective actions. Federal agencies review subsamples of the state samples to compute overpayment error rates using a regression estimator. These federal estimates make state-to-state comparisons of quality possible and are also used for fiscal sanctions. Such applications of the regression estimator have been widely challenged. We show that the regression estimates and estimates of their variances are closely unbiased, and we examine their statistical characteristics. We also comment on the recent reports of the National Research Council Panel on Quality Control of Family Assistance Programs.

Variance Estimation for a Specified Multiple-Frame Survey Design

This paper is concerned with the estimation of the variance of estimates based on a multiple-frame sample design in which alphabetic clusters of elements have been sampled from different lists, and where elements selected from different lists may correspond to the same element of the population of interest. A nesting procedure is used to assure that clusters selected from any one of the lists are also in the sample for any other list in which an equal or greater sampling rate has been specified.