How to Build a Survey – Branching, Skip Patterns and Randomization

Question Sequence When Building A Survey

The main tools used in building a branching survey are skip patterns, branching and randomization. 

A survey that allows different respondents to receive a different set of questions may be a result of the following:

Simple skip patterns:  The respondent answered one question in a specific way and therefore skips to a designated question further down the survey.

Branching:  A compound list of conditions is met (they answered several and met several different conditions) and were sent to a designated question further down the survey, or were only presented with a given question if conditions based on previous answers were met.

Blocking:  A set of questions are defined as a block and based on conditions met, the block is presented.

Randomization:  Used to prevent order bias in presentation of

• The order of question answers
• The order of questions within blocks
• The order of blocks of questions
• The random selection of a subset of questions with a block of questions

For example, a patron of a restaurant may be asked to evaluate the “overall quality of service received during their visit”.

If they respond that they are "very or extremely satisfied", we want to branch to ask the question "What was it about this visit that made it so enjoyable to you?"

However, if they indicate that the visit was less than satisfactory, we want to ask the question "What could have been done to make your visit more enjoyable?"

Branching is accomplished using Boolean logic statements of the form:

If (question # ) (state condition:   <,  = , >,  <=,  >=, not equal) (value), then (action: skip forward to the target question )
For example, If the answer on question 3 equals 1, then skip to question 5, Otherwise continue to question 4.

Survey branching must be carefully charted out. For simplicity, make sure all of your questions are in their final order before the branching pattern is implemented. It is best to flowchart questions and order questions so that you can visually see where your branching occurs and so that you will avoid logic and branching errors.

Question Sequence and Branching:

Conditional Branching occurs only if the condition is met. An example of conditional branching is the statement:
"If the answer to question (3) is greater than 2, then Branch to Question (5)"

      Compound Conditional Branching (compound branching means that branching is based on multiple answers, i.e., if question 3 is greater than 2 and question 1 is greater than 1). This option can be specified as a single instruction, or can be developed through multiple questions are included in the instruction set that defines the branching operations.

A branching example using a series of simple conditional branches might appear as follows: 

"Branch to Question (5) only if: " the answer to Question (3) is, as specified by the relational operator, greater than or equal to (4)"
AND a previous question had a separate branch instruction that directed the respondent to Question 5

Unconditional Branching occurs as a direct statement with no conditions. For example, "Branch to question 5" is an unconditional statement. Unconditional statements can be inserted for text questions or at the end of a branch path to bring the respondent to a specific point in the main survey.

Unconditional branching is the simplest form of branching, but can be used as a conditional branch if implemented correctly. To accomplish this, use unconditional branches for each answer of each question. As example, we might want to skip to question 15 if the respondent is a renter, but answer question 14 for owners.

            Caveats about Building Branching Surveys: Branching creates what are called "Opportunities for Disruption" meaning that the respondent often justifies discontinuing of the survey when a whole new page of questions appears. One recent study resulted in a 25% respondent discontinuation when the survey branched to a page with a large number of questions on it. Respondent termination occurs most often when the survey task continues beyond a reasonable time, number of pages, questions, or when they are not committed to the survey (or group sponsoring the survey).