A poorly designed questionnaire can be the biggest source of non-sampling error (either directly or indirectly). The questionnaire can influence the response rate achieved in the survey, the quality of responses obtained and consequently the conclusions drawn from the survey results.
Questionnaire design should be started by considering the objectives of the survey and the required output, and then devising a list of questions to accurately obtain this information. Careful consideration should be given to a number of factors including the types of questions to be asked, the questionnaire wording, the structure and design of the questionnaire and testing the questionnaire to ensure that quality data is collected.
This section discusses the developmental phase of questionnaire design, question wording, different types of questions and the structure and design of the questionnaire. Testing the draft questionnaire is covered in Survey Testing.
Before a questionnaire can be designed, time must first be spent clarifying the objectives of the survey with the client, researching the subject matter and methodology of the collection and conducting testing to obtain an insight into the characteristics of different groups of the target population.
This development phase needs to be undertaken to ensure that the data items in the final questionnaire are appropriate and can be collected accurately. It is also important to consider early on, the coding of questionnaire responses and the type of processing system to be used to process the questionnaire.
Defining Collection Objectives with the Client
The Set-up State of a Survey outlined the importance of clearly establishing the research objectives for the collection of any information. Similarly, in the process of questionnaire development, the reasons for collecting the information need to be clearly articulated. Defining the collection objectives with the client includes:
- clarifying the objectives of the survey;
- justifying the collection in terms of the benefits in the collection and its public good;
- determining the scope of the survey, ie who is to be surveyed;
- determining the desired output, ie specify the tables according to the objectives and the precision required;
- preparing a list of content, ie information needed according to the objectives. A number of questions may be necessary to obtain the desired information;
- defining the content;
- justifying the content, ie is all the information collected necessary; and
- establishing priorities for each data item. This is important in ensuring that the most important data items are collected. It also makes it easier to discard less important questions if the survey budget is reduced.
A similar process of defining and justifying the content in consultation with users should be conducted when adding new or redeveloped questions to an existing collection.
Researching the Topic
As stated in The Set-up Stage of a Survey, it is very important that the researcher who is going to design the questionnaire undertakes background research into the topic under consideration. In terms of questionnaire design, the research should aim to:
- clearly define the concepts and definitions to be used in the survey. The use of standard definitions and classifications will enhance data comparability; and
- accurately identify the target population characteristics. This needs to be done so that questions can be pitched at an appropriate level.
It is also worthwhile looking at past collections of the same topic to learn from past experience. Such collections can prove to be a useful basis for the current collection and also help to avoid possible mistakes. It is important to look at the concepts, definitions and question wording of past collections if a time series is to be created, since changes in these will result in changes in responses. For example, a question in the 1981 Population Census, 'Do you have a mortgage?' received very different responses from those that were obtained in the 1986 Census when the question was changed to 'Have you paid off your house?' Many people who had second mortgages included this in the 1981 Census but not in the 1986 Census.
Past collections of similar topics may also be worth looking at. For example, if you are collecting information on drug use it may be appropriate to follow a similar methodology to that used in surveys of alcohol use.
In examining past collections the method of collection used should also be considered. Some topics will be more or less appropriate for particular method.
The main functions of questionnaires are to extract information from respondents, record and process it. To achieve this effectively, a questionnaire should:
- provide clear instructions (for respondents, interviewers and processors);
- clearly and concisely define what is to be collected and recorded;
- maintain respondents' cooperation and involvement;
- enable respondents to complete it accurately and within a reasonable time;
- use a language that is understood by the respondents;
- avoid bias in question wording;
- make the job of the respondent and/or interviewer easy;
- appear uncluttered on the form;
- provide suitable space for responses; and
- be easily processed by both people and machines.
- be in a suitable form for keeping as a hard copy record.
Additionally, it is necessary to realise that a questionnaire often cannot provide all the information that a client would like to obtain. If a questionnaire becomes too long or confusing, respondents may be unwilling to complete it, or they may make mistakes.
TYPES OF QUESTIONS
Questions can be classified into types based on their answer formats and also the type of data sought.
Questions can generally be classified as one of two types - open or closed - depending on the amount of freedom allowed in answering the question. When choosing the type of questions, consideration should be given to factors such as the kind of information which is sought, ease of processing, and the availability of the resources of time, money, and personnel.
Open questions allow the respondents to answer the question in their own words. An example is 'What is your occupation?' The advantages of these type of questions is that they allow many possible answers and they can collect exact values from a wide range of possible values. However, they are more demanding than closed questions, both to answer and process. Open questions are often used in pilot tests to determine the range of likely responses.
Closed questions provide respondents with a range of the most likely answers to choose from. These questions are appropriate when the researcher can anticipate most of the responses and when exact values are not needed. However, they require more effort than open questions in the development and testing stages. Processing time of closed responses is much less than that of open ended responses. Some examples of types of closed questions are given below:
- Limited Choice
- Multiple Choice
- Partially Closed
Limited choice questions are those which require a respondent to choose one of two mutually exclusive answers. For example, yes/no.
These are questions where the respondent is required to choose from a number of responses provided.
Checklist questions allow a respondent to choose more than one of the responses provided.
These questions provide a set of responses where the last alternative is 'Other, please specify'. Partially closed questions are useful when it is difficult or impractical to list all possible choices.
Choosing Between Questions Types
In choosing between these two alternatives (open or closed questions), consideration should be given to factors such as the data requirement, the kind of information sought, the level of accuracy needed, ease of processing and the availability of coding resources, the position of the questions on the form and the sensitivity of the question. In general, closed questions are better for both the interviewer and the coder.
Once the questions have been chosen, they should be tested and retested until the best choice has been made. This is covered in more detail in Survey Testing.
Type of Data Sought
There are six main types of information or data that can be obtained from questions. They are discussed below.
In these questions, factual information is required of the respondent rather than an opinion. Respondents could be asked about possession of items (eg. 'Do you have a driver's licence?') or characteristics of the business (e.g. 'How many employees does this business have?').
Opinion or Motivational Questions
These questions seek opinions ('Are you in favour of a capital gains tax?') rather than facts. There are many problems associated with opinion questions. For instance:
- a person's attitude to a subject may not be fully developed or they may not have given it much thought;
- opinion questions are very sensitive to changes in wording; and
- it is impossible to check the validity of responses to opinion questions.
These questions require information about the activity of the respondent or business (e.g. 'How many times did you go to the theatre in the last 12 months?'). Behavioural questions need to be used with care because they often require difficult recall tasks on the part of the respondent. They should be restricted to topics respondents will remember easily or are likely to have records for, and cover a reasonable and specific time frame.
The "What would you do if ... ?" type of question. The problems with hypothetical questions are similar to opinion questions. You can never be certain how valid any answer to a hypothetical question is likely to be. They should only be used to refer to a hypothetical occurrence of a type of situation a respondent will be familiar with or have structure in place for. An example of this is for testing questions e.g. 'If an employee went on holidays at the end of January and was paid in advance for all of February, would you include them in the number of employees reported for the pay period ending on or before 21 February?'
Classification or Demographic Questions
These are used to distinguish the main groups of respondents in a survey for later analysis (eg. age, sex, industry). They are usually left towards the end of a questionnaire unless they are necessary for filter questions (questions which direct respondents to skip questions that do not apply to them).
These questions test the respondent's knowledge about current issues etc. For example, 'Who is the Prime Minister?', 'Are you aware of these industry support groups?'
QUESTION WORDING AND RESPONSE CATEGORIES
There are a number of factors to consider when designing questions to ensure that appropriate answers are obtained. Several aspects of question design can introduce error, namely:
Questions which employ complex or technical language or jargon can confuse or irritate respondents. In the case of interviewer based surveys, respondents who do not understand the question may be unwilling to appear ignorant by asking the interviewer to explain the questions. The respondent may then either refuse to answer or give an inaccurate response.
Technical language or jargon should only be used in cases where it is part of the normal language of the survey's target population. An example of this case would be a survey of information technology specialists: the survey would need to use language that is 'jargon' to the survey designer, but appropriate for the respondent.
A general principle to keep in mind is that the wording of questionnaire items should be specific, definitive, consistent, brief, simple and self-explanatory.
If ambiguous words or phrases are included in a question, the meaning may be interpreted differently by different people. This will introduce errors in the data since different respondents will be virtually answering different questions.
For example, consider the question 'Has your standard of living decreased substantially because of a sharp increase in your monthly mortgage repayments?' A 'No' answer could mean any one of a number of things - for instance: 'No my standard of living has not dropped because of increased repayments' or 'No, my repayments have not increased'.
A question may also seem straightforward, but allow for a variety of different kinds of answers. It is important to include the measurement unit you require wherever one applies, e.g. dollars, days, litres.
Double-Barrelled Questions (Multiple Concepts in one Question)
These are apparently single questions which actually incorporate two different questions. For example: ' Do you intend to leave work and return to full-time study this year?' A person may be intending to leave work, but not to return to study, and vice versa. When different parts of the question have different answers, or parts of the question are not relevant, respondents may be unsure how to answer. When attempting to interpret answers to such questions, it can be unclear to which part of the question the answer corresponds.
Error will be introduced if questions lead respondents towards a particular response. For example, the question 'How many days did you work last week?', if asked without first determining whether respondents did in fact work in the previous week, is a leading question. It implies that the person would have or should have been at work. Respondents may answer incorrectly to avoid telling the interviewer that they were not working.
Another form of leading questions are unbalanced questions. For example, 'Are you in favour of gun control?' provides only one alternative to consider. The question should be reworded to something like 'Do you favour gun control, or are you against gun control?', which gives respondents more than one alternative. The answer options of a question can also be unbalanced. For example, a respondent could be asked in a neutral way "Please rate your overall health" but required to select from the answers "Poor", "Good" and "Excellent".
A significant degree of error can be introduced in questions which require respondents to recall events, expenditure etc, particularly if details are being sought for a long period. The quality of the data collected from recall questions is influenced by the importance of the event to the respondent and the length of time since the event took place. Respondents also tend to remember what should have been done rather than what was done.
Subjects which are of greater importance or interest to respondents, or events which happen infrequently, will be remembered over longer periods and more accurately. Where possible (eg. with financial information), questions should be framed so that respondents can refer to their own records which would enhance accurate reporting. Minimising the recall period also helps to reduce memory bias.
A specific type of memory error is telescoping. This occurs if respondents report events as occurring either earlier or later than they actually occur, incorrectly bringing events into the reference period. This effect is alleviated somewhat by being very specific about when the reference period begins and ends, for example using "the week ending Saturday 1st September" rather than "last week".
Intrusive (Sensitive) Questions
Questions on topics which respondents may see as embarrassing or highly sensitive can produce inaccurate answers. Respondents may refuse to provide information on personal issues such as health or income details. If respondents are required to answer questions with information that might seem socially undesirable, they may provide the interviewer with responses they believe are more 'acceptable'. In these cases it is often better to provide the respondent with a self-administered questionnaire that the interviewer doesn't see.
Business survey respondents can also find some topics sensitive, such as IT security breaches or donations to charity, as well as not wanting to reveal commercial-in-confidence information about their business. Business surveys add a new dimension to collecting sensitive data as it is often necessary for different respondents, sometimes from different areas, to complete parts of the form and approve its content. Some respondents might not want the others to see particular answers.
The negative effect of sensitive questions may be aggravated if they are placed at the beginning of the questionnaire and can therefore contribute to non-response if respondents are unwilling to continue with the remaining questions. If a sensitive question is further into a form the respondent is more committed to completion, and if they do refuse to continue, the partial response is more useful. Ways of overcoming difficulties associated with sensitive questions may include reassuring respondents that the information they provide is confidential, and not requiring respondents to write their name anywhere on the survey form.
These questions seek to locate a respondent's opinion on a rating scale with a limited number of points. For example, a five point scale measures strong and weak attitudes, ie respondents may be asked whether they strongly agree/agree /neither agree nor disagree/disagree/strongly disagree with a given statement. Whereas a three point scale would only measure whether they agree, disagree, or neither, but not the strength. Using many scale points will generally not provide meaningful differences in attitude strength.
Care needs to be taken with the use of attitudinal scales because some respondents may have difficulty interpreting the scale. Additionally, such scales are interpreted subjectively and this interpretation can differ between respondents.
This situation arises when respondents have a general tendency to agree rather than disagree with anything. It occurs when respondents are asked whether they agree or disagree with a statement, especially when the supplied statements are presented as plausible generalities. It can also appear for questions requiring a yes or no response.
This tendency can be due to a combination of factors, such as the personality and education level of respondent, as well as conditions of the interview or design of a self-completed questionnaire. Respondents will often agree when the question is ambiguous or otherwise difficult to answer. The effect may be exaggerated when the respondent is fatigued or has to answer a long string of questions with the same response categories. A related effect is satisficing, where respondents select the first reasonable answer rather than make the effort to find or remember the best answer.
Adequate Response Categories
It is important to make sure that there are adequate response categories and that they incorporate every possible response. For example:
- Age 15-19 21-25
Another problem that could arise is overlapping response categories. Ranges should always be mutually exclusive. For example:
- Age 15-20 20-25
Response categories also need to be worded carefully, as respondents will use them to clarify or extend the meaning of the question. For example, if a question used a frequency scale with five points "Never", "Rarely", "Average", "Often", and "Frequently" a respondent may incorrectly assume the scale represents the population distribution. If they consider themselves to be normal or extreme compared to the population on the activity of interest their answers will differ regardless of the actual frequency they engage in the activity.
Number of Response Options
The number of response categories can influence the quality of the data as both too few and too many categories can cause errors. Too many can cause respondent fatigue and inattention, resulting in ill-considered answers. If there are too few categories respondents may have difficulty finding one which accurately describes their situation.
Don't Know Category
The decision about whether to include or exclude a Don't Know option depends to a large extent on the subject matter. The remaining responses are usually evenly distributed on the negative and positive sides of a scale if this category is excluded, although it depends to a large extent on the nature of the question. Excluding this option may not be a good idea, as respondents may be forced to give an answer when, for example, they really do not know what their attitude is to a particular subject, or they do not know the answer to a factual question that has been asked. When respondents are forced to develop an attitude on the spot this attitude will be highly unreliable.
A change in wording can result in a change in responses. For example, different response may be obtained through using the following two questions:
'Do you think that gun ownership should be forbidden?' or
'Do you think that gun ownership should not be allowed?'
Minor changes in wording can also have a significant affect on responses. One should therefore be careful when looking at alternative wordings. The use of negative words like "not" should be avoided in questions as they are easily missed by respondents. In addition, using "not" in a scale such as "Satisfied", "Neither" and "Not satisfied" doesn't provide a true opposite. "Dissatisfied" would be a better alternative, however "Unsatisfied" could also be used and would mean something slightly different to respondents.
STRUCTURE AND LAYOUT OF THE QUESTIONNAIRE
Not only does the wording of questions require attention to detail, but also the 'look' of the questionnaire. Poorly designed questionnaires (eg. hard to read text) not only serve as a disincentive to respondents completing the questionnaire, but can also result in respondents making errors. Some of the important elements of questionnaire structure and layout are outlined below.
The questions on a form should follow a sequence that is logical to the respondents. Regardless of the method used to administer the questionnaire, the sequence should flow smoothly from one question to the next. A smooth progression through the questions is particularly important if the questionnaire is answered in difficult circumstances (e.g. a mother trying to fill in a questionnaire while her children are seeking her attention). It is a good idea to start the questionnaire with pleasant and easy questions to promote interest in the survey, and to give respondents confidence in their ability to answer the remaining questions. In particular, the opening questions should establish that the respondent is a member of the survey population.
The remaining questions should be logically structured so that the interviewer or respondent does not need to alternate between pages of the questionnaire. For example, any explanatory notes should be presented as part of the question they refer to, not on a separate page.
Questions which may be sensitive to respondents should generally not be placed at the beginning of a questionnaire. Rather, they should be placed in a section of the form where they are most meaningful to the context of other questions. In this way the format of the questionnaire can act as a buffer to help the respondent feel more comfortable with sensitive questions after establishing rapport.
In self-enumeration questionnaires, to ensure that respondents answer only those parts of the questionnaire that are relevant, filter questions may be used to direct respondents to skip the questions that do not apply to them. Filter questions are also used in interviewer based surveys to direct interviewers to follow a series of questions according to answers given by respondents. Filter questions need to be used with care as respondents (and interviewers) need to have sufficient information about the skip condition to judge whether the respondent should skip. Filters should also generally be avoided for sensitive topics as respondents will tend to give the answer that avoids answering the sensitive questions.
If the instructions are not clear and straightforward, interviewers or respondents can follow an incorrect sequence or miss questions. In general, only one or two conditions should be placed in each sequence guide. Computer-assisted interviewing and electronic self-completion forms can make complex sequencing much easier.
Filter questions also identify sub-populations. For example:
Q7 'Were you born overseas?'
If 'Yes' go to Q8,
if 'No' go to Q12.
Order of Questions
The order in which the questions appear may influence the responses given to particular questions. Responses given to earlier questions can influence responses to later questions. For example, if a question asks participants whether they believe trade unions are disruptive in the community, and then a later question asks about problems in Australian industry, the negative influence of trade unions on industry may receive much more attention than would otherwise have been the case.
Order of Response Options
The actual order of response options can also introduce bias. The options presented first may be selected because they make an initial impact on respondents, or because respondents lose concentration and do not hear or read the remaining options. Equally, the last options may be chosen because they are more easily recalled, particularly if respondents are given a long list of options. Thus, the order of response options has a greater effect on data quality when a question includes a large number of response options.
If possible, options should be presented in a meaningful order. If some options are more socially desirable than others these should go last to reduce bias. For example, an education question should present the qualifications in order from lowest to highest. For some self-completed lists, alphabetical order is the most appropriate to help the respondent find which option they want, for example if respondents have to select which crops they produce.
Response Options and Respondent Difficulties
When the survey is interviewer based, the response options can be presented either verbally or on a prompt card. A prompt card is a list of possible responses to a question which are shown by the interviewer to assist the respondents. This helps to decrease error resulting from respondents being unable to remember all the options read out to them. However, respondents with poor eyesight, migrants with limited English, or adults with literacy problems will experience difficulties in answering accurately.
The length of a questionnaire can be described in different ways. Survey designers tend to worry about number of pages, whereas the number of questions (especially mandatory ones) and the time taken to complete are usually more important. How long is too long varies across mode of data collection, the ease and interest of the topic and the design of the questionnaire. Towards the end of a long questionnaire, respondents may give less thought to their answers and concentrate less on the instructions and questions, thereby decreasing the accuracy of the information they provide. If a respondent is told an interview will last several hours, or they receive a questionnaire that is many pages thick, that can lead the respondent to refuse to participate at all.
Respondents or interviewers using a questionnaire with poor layout can miss questions, follow an incorrect sequence or enter responses in the wrong response box, which will result in missing or incorrect data. Poor layout can also contribute to errors at the processing stage. For example, if response boxes are not aligned some answers may be missed completely during the data entry.
Particularly for those questionnaires which are completed by respondents, the design may contribute to errors as a result of:
- • poor legibility (e.g. unclear printing or very small text);
• violating the normal reading path of the respondent (English readers expect to read from top left to bottom right and e.g. big headings and bright pictures in the middle of pages disrupt this);
• instructions which can be easily overlooked (e.g. those which are not clearly differentiated from the questions or are not placed near the relevant part of the question); and
• inadequate space for answers.
The questionnaire should be physically set out so as to minimise the time needed to interview, respond and process the results. Specifically, consideration should be given to the form's construction, graphics, and layout. Poor layout leads to mistakes in understanding questions and recording the responses. In general, the questionnaire:
- • needs to be understood by respondents, interviewers and processors. This is done by providing clear instructions. For respondents this means adequate layout of questions and response categories with appropriate sequencing. Interviewers require prompts to be clearly understood and response codes need to be adequate for processors;
• should have a good appearance as it might affect the response, i.e. the questionnaire should be well designed and presented and therefore easy to answer;
• should clearly identify the date, title and the organisation;
• should clearly outline the purpose of the survey;
• should assure respondents about the confidentiality of the information they are providing;
• should provide a contact number so that respondents can obtain help if they require it and a due date; and
• should have pages that are numbered consecutively, with a simple numbering system.
Specific principles of good questionnaire design are outlined below:
- • left align text where possible;
• typeface: avoid ornate and decorative typefaces. Serif type fonts (eg. Times) are easier to read for questions (on paper) than sans serif types (eg. Helvetica);
• upper case text is difficult to read, and should be avoided where possible;
• allow enough space for answers;
• tick boxes are a popular way of obtaining responses. Where possible, they should be vertically aligned rather than horizontally aligned; and
• lines are sometimes useful in order to divide columns, sections and questions. These lines should be as fine as possible and they should only be used where necessary;
• keep the amount of ink on the form to the minimum necessary for the form to work properly. Cluttered forms contribute to respondent fatigue and errors, thus leading to a decline in data quality;
• black text is easiest to read;
• the background colour of the questionnaire form should not be too strong or bright. Light pastel colours are usually suitable. Background colour should contrast enough with the text to allow for ease of reading;
• avoid colour combinations such as red and blue or green and orange. Black on pale yellow is a good combination, but it is difficult to get the right yellow.
Once the understanding of concepts and definitions has been investigated through focus groups, a rough questionnaire can be produced and tested informally on a small group of people, perhaps one's colleagues at work. Such testing is not intended to obtain representative results, but aims to find out the major flaws with the questions, for example awkward wording. This testing is designed to take the 'rough edges' off a questionnaire. It is a good idea to use open questions to work out the likely responses. The questions can be restructured and developed into a draft questionnaire which can be used in rounds of informal pretesting and later pilot testing. For further details Survey Testing.
Questionnaire design begins by clarifying the objectives of the survey, determining the data which is to be produced by the survey and devising a list of questions to obtain this data. Careful consideration should be given to a number of factors, including the type of questions to be used, the logical sequence and wording of questions, and the physical design of the form. It is important to test each of these aspects of questionnaire design with a group of respondents before finalising the questionnaire. If necessary, the form can then be modified and retested until respondents can complete it accurately and quickly with a minimum of errors.