Customer Service: +44 1334 657957


Setting up repeat questions in Companion

There are a number of ways to handle repeated blocks of questions.  These methods are relevant whenever a respondent is asked the same set of questions for two or more products or situations. They may be used for small blocks of one or two questions, up to very large repeats making up the bulk of the questionnaire.

When deciding which method to use a primary consideration should be the accuracy of collecting the data. For this reason we recommend, whenever practicable, to enter the data directly from the questionnaires in the order which the questions appear on the printed document. We would particularly discourage any attempt to collate parts of individual questionnaires into some predefined order.

The descriptions which follow refer to a simple product test consisting of two repeats of two questions:

QA

Q1P

Q1L

Q1R

Q2P

Q2L

Q2R

QP

What age are you? (Young, Old)

Which product did you try first? (list of possible products)

What did you like about the first product? (list of likes)

Please rate the first product (Liked, Indifferent, Disliked)

Which product did you try second? (list of possible products)

What did you like about the second product? (list of likes)

Please rate the second product (Liked, Indifferent, Disliked)

Which did you prefer? (preferred first tested, preferred second tested, no preference)

To keep the analysis as simple as possible it is important that the responses for each repeated set of questions are exactly the same. So in our example, Q1P has the same responses as Q2P, and Q1L as Q2L and Q1R as Q2R. The entries QA and QP are only asked once.

We recommend using “similar” names for “similar” questions, as we have done above, rather than simply numbering them Q1 to Q8. It is much easier to produce the correct analysis when consistent numbering patterns are used from one repeat to the next.

One area of confusion that sometimes occurs when handling repeat questions is the base for individual tables. It is crucial that the person requesting a table, and the person producing it, both have a clear idea of the base for the table. In our example above we will assume that we have 100 respondents, who all try two products. We then have two possible bases for our tables:

Respondent based tables with a base of 100.

Product based tables with a base of 200 (or less if not everyone tests two products).

It is possible to produce tables for all of the questions based on respondents (100) or products (200). Sometimes one base will make more sense than the other, and sometimes one base will be easier to produce, but there is neither a correct base nor an incorrect one, see Repeats introduction.

Generally the rows down the side of a table will be used to determine the base.

For our survey, a typical respondent based table rows might be:

  • QA Age
  • V1 Products tested
  • VP Preference

Q1P and Q2P cannot be used directly on respondent based tables because there are two of them for each respondent. Therefore it is necessary to make a multi variable (V1) using the block insert selection, which includes both questions combined with “or” so that each respondent will then have two products listed. Variable V1 gives the answer to the (not asked) question “Which two products did you try?”

A variable V2 could be created from Q1L and Q2L in the same way, giving the likes for the respondent. This variable would answer the (not asked) question “What did you like about either or both of these products?”. Variable V2, however, is not useful because you do not know which “likes” refer to which product. It is a common mistake to tabulate V2 by V1 on a respondent based table – the table will be produced and will be “correct” but useless.

Tabulating QP directly also is not helpful, so a variable VP should be made with the responses “Preferred product A, Preferred product B, No preference”. If there are four products being tested, but each respondent only tries two of them, VP will need four product preferred rows, plus a “No preference”. VP is not difficult to define – each row is taken in turn and defined from any of the relevant locations. For example “Preferred product A” is either “Preferred first” and A tried first, or “Preferred second” and A tried second. If A is Q1P/1 and Q2P/1 then the definition for “Preferred A” would be (Q1P/1 and QP/1) or (Q2P/1 and QP/2).

For our survey, typical product based tables might be:

  • Q1L + Q2L Likes
  • Q1R + Q2R Ratings

As we have seen above, making a respondent based variable for these questions is not helpful. To analyse these questions it is necessary to produce product based tables. This is done by producing two separate tables (one for first tried and one for second tried) and adding them together, using the overlay table facility.

A typical table would be Q1L by Q1P and overlaid on it Q2L by Q2P.