What is coding?

Surveys nearly always have closed questions and often have open questions.

 

Closed questions are those that require the participant to, for example, tick boxes or score on a scale to best represent their views. These questions can be easily analysed from within the survey application.

 

Open questions allow participants to respond in their own words, which can generate hundreds or thousands of sometimes lengthy and detailed comments; the real meat on the bones is often revealed here, sometimes with unexpected results. Since these comments tend to relate to prior questions in the survey (for example, 'why did you give that score?') it is helpful to analyse them alongside the closed question data, presenting the fuller picture, in context. 

​

Before any analysis can be performed, open question responses need to be coded. Coding is a specialist technique consisting of the assignment of numerical codes to recurring words, phrases, comments, suggestions, opinions and feelings.

 

Whilst coding of open question comments might be viewed as a simple yet laborious process, the importance of it should not be underestimated; it requires skill and expertise to make it worthwhile.

​

The coding process:

​

• Listings created: usually 10 to 20% of comments are thoroughly read and a list of recurring responses tallied.

​​

• Code frames built: each recurring comment / theme from the listing is assigned a unique numerical code. Codes are usually assigned where a comment occurs 5 times or more within the sample (though naturally dependent on sample size). The codes and corresponding comments (code labels) collectively form the code frame. 

​​

• Coding performed: each and every comment is read and understood and the relevant code(s) assigned to each comment. Where a comment doesn't fit with any of the codes, an 'other' code is assigned, and these can then either be recoded into broader categories or left for later reference.

​​

• Coding verification: coded comments need to be verified to check for accuracy of coding. 10% back-checking is usually sufficient, but if budget allows then as much as 100% back-checking can be performed. Additional checks include running counts on the codes (how many times a code has been used), checking there are no counts in blank code labels, checking the 'others' to see if they can be recoded, and general sense checking of code labels. 

​​

• Coding analysed: codes, along with case IDs, are then imported or manually entered into the analysis application and analysis performed as required. 

​