International Handbook of Survey Methodology. FULL ACCESS. Full Access: You Imprint Routledge DownloadPDF MB Read online. lesforgesdessalles.info: International Handbook of Survey Methodology (European Association of Methodology Series) (): Edith D. De Leeuw, Joop Hox. The International Handbook of Survey Methodology opens with the foundations of survey design, ranging from sources of error, to ethical issues. This is followed .
|Language:||English, Spanish, Portuguese|
|ePub File Size:||27.40 MB|
|PDF File Size:||14.85 MB|
|Distribution:||Free* [*Regsitration Required]|
International Handbook of Survey Methodology. Edited by. Edith D. de Leeuw. Utrecht University. Joop J. Hox. Utrecht University. Don A. Dillman. Washington. International Handbook of Survey Methodology. Edith D. de Leeuw, Joop This Document PDF may be used for research, teaching and private study purposes. Request PDF on ResearchGate | On Jan 1, , E.D. de Leeuw and others published International Handbook Of Survey Methodology ().
Intended for advanced students and researchers in the behavioral, social, and health sciences, this "must have" resource will appeal to those interested in conducting or using survey data from anywhere in the world, especially those interested in comparing results across countries. With such surveys. Measurement error can be prevented by asking clear questions. In the Philippines. Although attention to the theoretical principles summarized earlier can help researchers to attenuate context effects in attitude measurement.
Some of these factors vary between countries and often lead to response rates differing between countries for the same survey. If we cannot successfully collect data from a large proportion of the selected units.
Chapter 6 allow us to make inferences about a population of interest. It may seem rather negative to be discussing nonresponse so early in this book. In this chapter. But this is a fundamental aspect of survey research. But whatever the circumstances of your survey. This requires advance planning—even before the sample has been selected. But it may also be different in other important ways that affect the estimates.
Even the most well resourced surveys carried out by experienced survey organizations suffer from nonresponse. This is rarely the case. I try to explain how and why nonresponse occurs.
The level of nonresponse can vary greatly between surveys. But nonresponse The principles of statistical inference see Lohr. This means that we can expect to find low ability persons in the responding sample of The probability of a particular unit being in our final responding sample. The result is that our sample may no longer be representative of the population. The sample design is to randomly select one in every 20 adults see Table 3. Suppose we want to estimate the proportion of adults classified as low ability.
Imagine that the population of Peter Lynn 36 disturbs the selection probabilities. In our example. If such. In the absence of other information. But if we were carrying out this survey for real. Suppose however that the low ability persons are less likely to respond to the survey. Table 3. Assuming that we have used a probability sampling design.
Figures in bold would be known. But nonresponse rarely happens completely at random. The effect of nonresponse on a survey of literacy High ability Population Low ability Total 8. Consider a simple example of a survey of literacy in a small town. If nonresponse had happened completely at random. There are reasons why some units do not respond and those reasons are typically associated with at least some of the survey variables. But the second is not known.
We would only observe the numbers highlighted in bold in Table 3. Inability of the data collector and sample unit to communicate e. Inability of the sample unit to participate e. Failure to make contact with the sample unit c. This may prove impossible if. These reflect the stages of the survey data collection process. There are several reasons why nonresponse occurs. Reasons for nonresponse a. If located successfully.
If we are to be successful in trying to minimize the extent of nonresponse. In particular. Once a sample unit is selected. Nonresponse 37 people were selected into the sample. There are also differences between surveys of individuals and households on the one hand and businesses and other establishments on the other.
Even if contact is made successfully. And if people with low literacy ability were more likely than those with high ability to engage in this seasonal employment.
In the case of individuals and households. A summary classification of reasons for nonresponse appears in Table 3. Refusal of the sample unit to participate d. Reasons for this can be broadly classified into three types: Similar designs can be found in several other countries. Surveys that use this design include the British Crime Survey. Let us consider some common types of survey. The first stage of the process is to mail an advance letter or prenotification letter to each selected address.
The sample is usually selected from a list of either persons or addresses e. The shaded boxes indicate nonresponse outcomes. But this is not always successful. This notifies the residents that an interviewer will be visiting soon. They must then make contact with the residents. These cases are typically referred to as office refusals.
At each stage. The fieldwork process is summarized in Figure 3. Having received this letter. To illustrate this. Where possible. The process for a sample of addresses. Not traced. Not yet built. Insufficient address. Laiho and Lynn On United Kingdom surveys of this kind. It can be seen that the survey participation process is quite complicated and there are many stages in the process at which there is an opportunity for nonresponse to occur.
In some such cases. Telephone surveys are also often used when the sample is of named persons for whom a telephone number is available. It may not be the person who provided the information to make the listing. In the majority of cases. Once the random selection of a person to interview has taken place. It is increasingly common in some countries for people to use devices that enable them to see the phone number of the person calling them before they answer the phone.
The selected person may be unable to participate due to illness or incapacity or may not speak adequately the language in which interviews are being conducted. It is much easier to refuse on the phone than to an interviewer standing at the door. This usually involves selecting a random sample of phone numbers by a method such as random digit dialling RDD. In some countries. If the phone number is correct. With such surveys. The selected person may refuse. And even if contact is made.
In the small minority of cases where an address contains multiple dwellings. They may choose not to answer if they do not recognize the number. If contact is successfully made.
Some people refuse to provide the information necessary to list the residents. Chapter 7. With this kind of survey. There are many reasons why this happens and.
Amongst cases where the questionnaire successfully reaches the sample member. In some cases this represents a refusal. For web surveys. The difference is that it is hidden from the view of the survey researcher to a greater extent than with interview surveys. And some questionnaires may be completed but get lost in the post. But in reality the underlying process is still quite complex.
This is where a respondent gets a certain way through the questionnaire and then decides not to continue. Nonresponse 41 3.
This tends to happen during periods of high flow. This can be caused by incorrect or out-of-date email addresses. For invitation-only surveys. There may be some cases where the respondent is unable to complete the questionnaire due to illness. Examples include international passenger surveys that sample and interview at ports and airports.
If you are attempting to interview people while they are waiting in a queue you may get rather low levels of refusal as the sample members do not have many alternative ways to spend the time. The ways in which the survey organization controls and facilitates that process are likely to influence the extent of nonresponse due to a failure to reach the relevant person s —a form of noncontact.
The behavior and performance of both the sample member and the interviewer during the interaction will be largely influenced by two sets of factors. The extent to which this happens depends on the frequency with which people are sampled at each sample location determined by the population flow and the sampling interval and the number of interviewers working at that location.
Peter Lynn 42 as interviewers are still occupied interviewing previously sampled person s.
Refusals often constitute a large proportion of survey nonresponse. Many business surveys are conducted as self-completion surveys. A conceptual framework for survey co-operation in the case of interview surveys is presented in Figure 3.
These can be broadly labeled the social environment and the survey design. Both actors in this interaction will of course also have their own personal characteristics and predispositions upon which these two sets of factors act. But if you are sampling people who have just disembarked from a train. The decision about whether or not to co-operate is an outcome of the interaction between interviewer and sample member. This raises a different set of concerns regarding confidentiality and sensitivity of responses.
The social environment includes the degree of social cohesion. These influence the degree of social responsibility felt by a sample person and the persuasion strategies and decision-making strategies used by interviewers and respondents respectively. The extent of refusals will largely depend on the time that sample members have available and the circumstances.
Adapted from Groves and Couper A conceptual framework for survey co-operation. These are discussed in section 3. Many aspects of survey design affect response rates. But if the interview is short. Mode of interview is very important. Relevant factors include comfort and perceived safety. These tactics may have different implications for the survey outcome.
How interviewers introduce the survey is also likely to be influenced by the length and content of the interview. They cannot show the sample member documents or identity cards. These limitations may contribute to the lower levels of success that interviewers seem to have in avoiding refusals on telephone surveys.
Social environment Survey design Sample member Interviewer Sample member Interviewer interaction Decision to co-operate or refuse Figure 3.
Interviewers are much more limited in the ways they can communicate with a sample member if they are talking on the telephone rather than standing in front of them face-to-face. Cialdini and Couper discuss six psychological principles that apply to requests to take part in surveys: Cognitive effort essentially relates to how difficult the questions are to answer.
The more. The framework presented in Figure 3. The survey researcher should therefore. This idea is nicely encapsulated in the leverage-saliency theory of survey participation Groves. The benefits and drawbacks will be weighed up against one another and if the drawbacks appear to weigh more heavily. The more relevant the survey appears.
Various materials are available to assist in training interviewers in techniques to maximize response rates. The interaction with the sample member now.
But being interviewed can also have negative consequences. This is perhaps one reason why self-completion surveys. Other aspects of burden include cognitive effort. For many people. Sensitivity refers to embarrassment. The interaction is therefore much more limited and the survey organization rarely has the opportunity to react to particular concerns or circumstances of sample members.
Every survey should document the outcome rates achieved.. Published response rates are often accepted uncritically. In the case of a postal survey.
Even more fundamental is the way in which the outcome categories themselves are defined. It can also be used as an indicator of the success of the data collection operation.
These rates should be calculated in clearly specified ways. This too should be documented explicitly. The guidelines referred to earlier provide a set of standard definitions of outcome categories that can be applied to most surveys.
This can make comparisons of published response rates fairly meaningless. We saw earlier in this chapter that there are many possible ways in which nonresponse can arise on a survey.
A single response rate does not convey that information—a complete distribution of outcomes is needed. In the case of a web survey. Strategies that can be adopted to minimize refusals on self-completion surveys are discussed in Dillman Nonresponse 45 typically consists of the sample member reading written material. No single. It is good practice to publish the number of sample cases in each outcome category e. If we want to learn how to improve response rates next time. In the case of face-to-face surveys.
Chapter 6. This may require interviewers to travel to different areas. The researcher should consider carefully how. This can partly be achieved by good training. With a more dispersed sample. I outline below some techniques that have been found to work well in some common survey situations. The researcher should consider at an early stage whether this is likely to be necessary and.
It is important to provide interviewers with motivation to make extra calls. Some population subgroups—for example. This is particularly true for face-to-face and telephone interview surveys. The classification in Table 3. Paying a fixed hourly rate provides no incentive for interviewers to call at times when.
With a clustered sample see Lohr. Interviewers can reduce noncontact rates by making more call attempts and by varying the times of day and days of the week of their call attempts. We need to combine many techniques. This presents challenges for at-home interview surveys.
During field work. The necessary extent of the efforts. Many survey organizations use computer assisted systems for telephone surveys. On a larger survey. The design of postal survey packages is discussed by de Leeuw in Chapter This requires a carefully planned call scheduling system. As some people can be away from home for long periods on holiday.
The appropriate strategy depends on the rate of flow. Once the mailing has arrived at the correct address. If self-completion questionnaires are to be posted to sample members. Even if no contact at all is made.
Paying a modest bonus for achieving a target contact rate could be effective. The marginal cost of making extra call attempts is relatively low on a telephone survey so many attempts can be made. If sample members are being telephoned at their homes. A plain envelope may be best. It may involve having. Nonresponse 47 people are more likely to be at home rather than times when they themselves prefer to work. The most important determinant of noncontact rate is therefore likely to be the quality of the address information used for the mailings.
All these counter measures are. For further discussion of making contact on web surveys. On a modest sized survey. The system should ensure that an interviewer it may not necessarily be the same interviewer calls back at an appropriate time if an appointment is made or if an indication is given of when the sample member is likely to be available.
It is not uncommon for survey organizations to stipulate that a sample telephone number must be attempted at least 12 or 15 times before it can be classified as a noncontact. On web surveys. The important thing is to ensure that field workers are able to deal adequately with periods of high flow.
If contact is made with someone other than the sample member. Surveys that aim to sample from a flow as described earlier are rather different from other surveys in terms of strategies to minimize noncontacts. If the work is being carried out from a telephone unit or other central office location. Sample members may be concerned that their answers should not become known to anyone else. It is good practice to notify the local police station in areas where you are carrying out in-home interviews.
Explain that results will be made available only in the form of statistical summaries—no individuals will be identified. Emphasize the aspects of the interview that people are more likely to find interesting. Be sure that interviewers carry identification and that sample members are given the name and telephone number of someone who can verify that the survey is genuine.
Interviewers can tell wary respondents that.
Some sample members may think that taking part will be too difficult for them. On an in-home interview survey. There is considerable experimental evidence that such incentives can reduce survey refusal rates. On interview surveys. Tell them that the survey is confidential and that nobody outside the research team will be able to link their answers to their name or address you must.
Offer to make an appointment. The survey should be introduced in a way that makes participation seem likely to be interesting and enjoyable. Tell them that they will not receive any direct mail as a result of taking part and that they will not be asked to take part in any further surveys if this is true.
Explain that the survey serves useful purposes. People might be more willing to take part at certain times than others. Tell them that you are interested in the views and experiences of all kinds of people—that the survey results must represent everyone. This should be minimized by keeping questionnaires as short as possible — ask only questions that are necessary.
Offering survey respondents a token of our appreciation helps to establish the bona fide nature of the survey and makes them feel better disposed to reciprocate by offering their co-operation in return. Provision of payment or a small gift can also help. Tell them that the questions are not difficult and that no specialist knowledge is required.
Be flexible and allow them to take part when it is most convenient for them. It is necessary to provide translated. The letter should also provide the name and phone number of someone to whom queries can be directed.
For many surveys. There is no point asking a proxy respondent about things that they do not know.
It is therefore important that interviewers are well trained in what to say to avoid getting a refusal. But including them is likely to be expensive. This can sometimes be appropriate. It is generally best to avoid mentioning how long the interview will take in the advance letter—leave this to the interviewer to explain. Depending on the nature of your sample. Other ways of reducing the number of temporarily absent sample members include extending the field work period and offering alternative modes of response.
On a postal survey. This person. And it is certainly not possible to ask opinions or attitudes by proxy. It should explain that an interviewer will be in touch shortly. Interviewers should be prepared to offer to come back when there will be someone else there too. You may also include a leaflet containing further information about the survey or about the organization for whom the survey is being carried out.
The letter should also briefly outline the nature of the survey and explain that answers will be treated confidentially. Excluding them would certainly introduce nonresponse bias. Sample members will decide. If the letter has an official letterhead. The method of communicating all these messages to sample members depends on the survey.
Nonresponse 49 the police know about the survey and suggest that they contact the police station to check this if they wish. In our earlier example. Peter Lynn 50 materials and. To estimate the extent of nonresponse error. And translation of survey materials is not a simple matter see Harkness. But we only observe y for the respondents in the sample.
Suppose we want to estimate a characteristic Y. This is the product of two components. We therefore need to pay attention to both these components. We estimate Y by the corresponding sample statistic y. We can express this as follows: This could be any kind of population parameter: The second.
The first. No qualifications But there are several possible approaches. It is a good idea to look at every available source of information about nonresponse as this helps you to build up a picture of the nature of nonresponse on your survey. If we include this information on the sample file. To illustrate the use of this expression for nonresponse error. Lynn Table 3. Standard grades only We could be fairly sure that nonresponse error would cause us to underestimate the proportion of young people who are unemployed at age This information has been used to derive an ordinal variable with eight categories.
Using sampling frame data thus has the advantage that nonresponse error can be calculated directly. The usefulness of the statistic lies in the fact that leaving school with very low qualifications is correlated with other parameters that we might wish to estimate using the survey data.
Because we know the level of qualification achieved by each sample member. Only rarely is this possible for individuals. The sampling frame for this survey includes a record of examination passes achieved at school. The response rate is highest amongst the most highly qualified sample members But linkage is often possible at some higher level of aggregation. The sample for a general population survey can be linked to such auxiliary data provided that suitable geographic identifiers exist on the sample file.
The data can then be used in the same way as for sampling frame data. A variation on interviewer observation is to collect data about nonrespondents by proxy. The data on these characteristics can then be used in the same way as for sampling frame or linked data. Some data items may be more sensitive than others to such differences. If these data relate to one or more of the same variables about which data have been collected by the survey. There may be differences in the time period to which they refer.
Other factors affecting the comparison include coverage error and sampling error. If you are planning an external comparison. You can then observe how these data relate to the survey variables to obtain an indication of the likely direction and magnitude of nonresponse bias. This can be enlightening. These factors are confounded.
In consequence. Process data of this kind. Nonresponse 53 is rarely very satisfactory as a means of studying nonresponse. The definition of the classes is therefore important. The advantage of this is that we typically have a rich range of variables available and at least some of them are likely to be highly correlated with the survey variables of interest.
Consider again the data of Table 3. This makes intuitive sense. For the first wave. Care is needed. Peter Lynn 54 3. This can be done quite simply using weighting. After weighting has been applied. But rather than simply describing it. But for subsequent waves. It can be seen that the error is now a weighted sum across the weighting classes of the difference in y between respondents and nonrespondents.
The response rate amongst sample members in category 1 was When choosing between. For nonresponse weighting to be successful. The weights will be greater the lower the response rate: A term applied to a number of post fieldwork procedures. Equally importantly. Tackling nonresponse involves carrying out every stage of the survey in a thoughtful. Data collection procedures should incorporate appropriate reminders or multiple attempts to contact sample members.
When designing field control documents and sample control systems. The difference between a survey estimate and the equivalent estimate that would have been obtained if all selected units had responded. In almost every chapter you will find references to it. Almost every stage of the survey design and implementation process has the potential to affect nonresponse error. An important point to remember at this stage is that it will not be possible to implement effective weighting unless you have planned ahead and collected some of the kinds of data outlined in the previous section.
When specifying the sample selection method. You do not currently have access to this article. Download all figures. Sign in. You could not be signed in. Sign In Forgot password?
Don't have an account? Sign in via your Institution Sign in. Purchase Subscription prices and ordering Short-term Access To purchase short term access, please sign in to your Oxford Academic account above.
View Metrics. Email alerts New issue alert. Advance article alerts. Article activity alert. Receive exclusive offers and updates from Oxford Academic. Related articles in Google Scholar. Citing articles via Google Scholar. On the Question of Generosity: Substance Abusing Inmates: Aside from the above, Dr. Gideon has published a methodology book titled Theories of Research Methodology: Respondents Cooperation: Comparison of Modes of Survey Administration.
Buy Hardcover. Buy Softcover.