Experiment with pattern-based auto coding

23 January 2014 - IN auto, coding, Datasets, pattern

Experiment with pattern-based auto coding

It’s not often that we get the chance to really come to know our data. Research opportunities such as that need to be savored. Far more often we are asked to generate results with a deadline of yesterday.

As a trainer I am often asked, “Can NVivo do the analysis for me?” At that point I think to myself:

I have some bad news and some good news. The bad news is that NVivo cannot do the analysis for you. The good news is that NVivo cannot do the analysis for you, so you’re still needed.

In deriving meaning from data the researcher will always play an important role. There are so many nuances in human thought and language that just cannot be adequately captured by an algorithm.

Sorting the data

That being said, there are some aspects of the process where an algorithm can do an initial sorting of the data so you, as the researcher, can focus on recognizing and organizing the more subtle differentiations. This can be extremely helpful when you have large volumes of data and a short period of time to go through it.

For example, you could use NVivo's auto coding features to organize a survey dataset—gathering all the responses to each question or gathering the responses by participant.

A new type of auto coding

Pattern based auto coding is an experimental feature designed to speed up the coding process for large volumes of textual data. It allows you to do approximate “broad-brush” coding of larger volumes of text quickly which you can then review and refine.

This type of auto coding looks at words within segments of selected sources and compares them with words within previously coded nodes. The scope is determined by selections in the Source List View and the auto-coded content is coded to the user profile “NVivo” (NV).

As the user, you can set it so NVivo will code more or less data, as well as select the granularity of coding from sentences to paragraphs to entire data cells.

For example:

For step-by-step instructions on using pattern-based coding, you can refer to Automatic coding using existing coding patterns.

Tips for success

There are steps you can take to get better results:

  • Consistent coding is helpful, irrelevant content may influence the results.
  • Code to descriptive rather than interpretive nodes.
  • The presence of advertising in your source materials can negatively impact your results.
  • The more you code in the beginning the better your results will be.
  • Having a variety of words in the nodes can help the algorithm better identify patterns.

We are excited about pattern-based auto coding and hope that you will take some time to try it out. Please share your feedback with us so we can continue to refine and improve this feature.