Text Questions

This page contains the following subsections:

• Introduction and general outline of the procedure
• Example
  • Step 1: Frequency table
  • Step 2: Removing capital characters
  • Step 3: Naming Categories
  • Step 4: Using Recode
  • Step 5: The first analysis
• Some final remarks about coding
• A playful alternative

Introduction 

Text questions are in general quite diverse. The questions we encounter can be split up into four main categories. Adding a fifth option "other kinds of text questions" could make this list complete.

1. Short open questions where the researcher expects a short answer. Examples are the spontaneous naming of a product or a typical property of a certain product.
   The question "Other, namely ..." also belongs to this category.

2. Somewhat longer open questions where the researcher asks for a short description.

3. An open question asking for comments in general. For example in a customer satisfaction survey one could start with a set of Likert scale questions and end with: "If you have any additional remarks or comments, please make them here."

4. Open questions in a qualitatitve research where a huge variety of answers is possible.
   
   and then there is category 5

5. Open questions that don't fit in one of the categories 1-4.

General outline of the procedure:

• The complexity of processing the answers increases from category 1 to 4. Processing answers from qualitative research is quite hard using SPSS 20. There is special software for that kind of analysis and one had better use it. See the Wikipedia article "Computer assisted qualitative data analysis software" for some options.
      

• Processing general remarks and comments usually will give a large variety of answers. In most cases it is sufficient to make a list of all the answers. You could for example use Analyze > Reports > Case Summaries.
  Read this list and use your common sense to find the major issues. Summarize them. You can cite one or two answers in your report. Maybe sort the list in some way and then put it in an appendix.
  
  
     

• The above categories 1 and 2 usually are categorized. Smart coding in SPSS 20 can help you to formulate and present the results. Depending on the number of answers and their veriety you choose one of three options. You code it as:
  • a qualitative variable;
  • A multiple response question with Dichotomies;
  • A multiple response question with Categories.

Example

Classifying answers to open questions into categories is time consuming, even with specialized software like SPSS Text Analytics.
We will show you in the example below how you can proceed using just SPSS Basic. We will use the file text_question_1.sav as our starting point.

Note: The answers in the data file stem from an online survey, so they have been entered by the respondents. If you have to enter the data into SPSS yourself (from paper forms) then of course the process will change, because some of the editing will no longer be necessary and you can browse through the given answers before the data editing and in this way make a preliminary list of answer options.
However, the basic idea of categorizing remains.

Step 1: Frequency Table

Start by making a frequency table or list with all the answers, using either Analyze > Descriptive Statistics > Frequencies or Analyze > Reports > Case Summaries:

From the output we infer that the 98 respondents gave a total of 45 different answers, where "different" only means that the string values are different. But for a human reader some "different" answers are the same, because:

• Due to capitals and lower cast letters strings are different;
• The same appliance is spelled differently: ipod, iPod and Ipod;

Furthermore:

• Some people gave more than one answer. This is due to a slightly ambiguous phrasing of the question. Apparently "Name an electronic appliance" doesn't stress enough that we want exactly one appliance.
  How to process these multiple response answers is up to the researcher. For example he or she could decide to use the first device mentioned and to ignore the rest of the answer.

We will continue by recoding the answers into a single qualitative variable.

Step 2: Removing capital characters

We can use the Compute command to do this (from the menu: Transform > Compute Variable...).
The screen dump below shows how it can be done using the string function Lower. We compute a new string variable V02 from the existing string variable V01.
Try it out yourself. Our intermediate result is the file text_question_2.sav.

This reduces the number of different answer options to 35. It is a first small improvement.
And it will pay off again in step 4, when we will use Recode.

Step 3: Naming Categories

The next step is transforming V02 into a numeric variable using Automatic Recode. The reason is that processing numeric variables is a lot faster and is better usable in SPSS. And you can easily combine numeric values into a group, while for strings that is not so easy.

Now study the LOG in the output.

After studying this list I came to the categories shown below. To implement this I will have to make a new variable (again) with these values and value labels.
Furthermore, I will have to recode the current variable V02_n into this new one.

Step 4: Using Recode

The answers for variable V02_n form the starting point for the creation of the new variable. We will call it "V01_categories". It helps the recoding to show it in a well-organized table.
You can also print the long list of answer codes from above and write the target value for the new recoded variable next to it.

Remember our earlier decision that we will only record the first answer given by a respondent. If you decide otherwise, then you will have to create four copies of the variable "V01_categories", all with the same 10 coded categories. The second, third and once even the fourth appliance mentioned will be recorded in the 2nd, 3th and 4th copy of V01_categories.

We have:

Implementing this in SPSS uses Recode into Different Variables (from the menu Transform > Recode into Different Variables...):

Once all recodings from old to new values have been defined, it looks like this:

Next we add the value labels to the newly created variable and we define 9 as its missing value:

This final data file can be found as text_question_4.sav.

Step 5: The first analysis 

Finally we can give an overview (tabular or graphical) of the answers to the open question we started with:

Some final remarks about coding

• As you can see all 98 respondents have their first answer coded. Nobody skipped the question.
• Two answers stand out. (Laptop / computer and (mobile) phone). All other options have been mentioned by just a few people. To summarize our findings we could combine them into a third group "other appliances".
• This coding process is quite time consuming. But if you want to make crosstabulations or compare groups in other ways, it is a necessary step. Whether it is worth the effort is to a large extend determined by the research questions that you formulated at the start of the research.
• If you want to avoid this coding process, then you have to think ahaead when you set up and pretest your questionnaire. Meaybe you should make the question more precise or give the respondents a predefined list of options. But of course that would change the nature of the question.
• Realize the difference between an open question with an answer field for a short answer (a spontaneous reaction by the respondent) and a question with a list of options given (a helped response). Which option is the best for your research depends on the research question(s) you hope to answer. And you have to balance the costs of the lenghty coding process versus the benefits of more spontaneous responses from your respondents.

A playful alternative.

We used the website: www.wordle.net to create the visual display of our table.
Note that in order to get it we had to adjust some of the labels.

The font size of each word is proportional to the frequency in the table above.

Please name an electronic appliance you couldn't live without.

Source: Research IB&C, 8 April 2010, by Jos Seegers.