NodeXL for Survey Data

Jan 21, 2013 at 5:32 AM

I'm curious whether anyone has used NodeXL for survey data? Survey researchers often use factor analysis, cluster analysis, etc. to show patterns, but results may be difficult to interpret for people without stats backgrounds. Would NodeXL conceivably work for survey data? 

For example, imagine a single check-all that apply question about individuals favorite types of NodeXL network analysis: Twitter, YouTube, Flickr, etc. Could the relationship between the respondent and their connection between these NodeXL network analysis be mapped to show how closely related these analysis are (e.g., if Twitter and Youtube are favorited more together relative to other analysis)? If so, would it be connecting respondent's to their relationship to the analysis or would it be linking each analysis to all of those it is mentioned with? I suspect the latter, but the data manipulation might be difficult to achieve. 

Coordinator
Jan 21, 2013 at 2:08 PM

Thank you for your interest in NodeXL!

Any set of connections can be analyzed and visualized in NodeXL, including respondent to question reply networks.

Today, you will need to create an Edge List from the survey to import the data into NodeXL.  In this edge list each Respondent has a relationship to each Question with a value equal to their reply.

The ONASurvey tool provides direct integration with NodeXL via GraphML import.  See: https://www.s2.onasurveys.com/

We do plan to import from two mode matrix in the future (though no date is set for that work item yet).

Regards,

Marc

Jan 21, 2013 at 7:06 PM
Hello Marc,

Thanks for your response. I've used NodeXL in the past for social network analysis, but have to confess my overriding interest is its potential application to surveys. I read your book, but it was a year or two ago, so I'm going more off of memory on how the algorithms work. My background is in survey research and I think the software could have a great capacity for helping decision-makers visualize data.

With all of that said, I think there are some easy ways to conform typical survey data to NodeXL formats. My question is more about the interpretation of the data. I've attached a few files to demonstrate the process that your typical user might go through in order to get an edge list. In fact, the steps I'm proposing are geared to be free of cost or at the lost possible cost in order to be accessible to your users who might have an interest.

First, collect survey data using an open source or subscription service. These are widely available and fairly cheap. Moreover, students can access them for free. The attached file represents randomly generated data (file 2) for a one-question survey (file 1) created through "Survey Gizmo" and exported to excel.

I then used an excel macro to flatten/reshape the data. These can probably be written pretty easily. I downloaded a free add-in from Tableau software. Afterwards, my data was in the correct format, with a row representing an edge between a respondent and their selection.

I then pasted the two columns into the Edges tab in the NodeXL add-in and generated a visual using Harel Koren (file 4). My question relates to the interpretation of this graph. Given the force-driven algorithm, can I assume that the reasons for using NodeXL ("reasons") are represented in terms of their relationship to one another on this graph? For example, since "some other reason" and "all of the above" are closer in proximity, does that then mean that they were chosen together as reasons more frequently relative to the other reasons? I'm guessing not necessary, since the graph maps out respondent relationships to each reason, and not relationships between the reasons directly.

The latter would involve an edge for each reason relationship for each respondent. For example, if a respondent selected "Academic" and "Professional" as there answer, there would be an edge between "academic" and "professional" in the worksheet, not the "respondent and academic" and the "respondent and professional."

Thanks and any thoughts greatly appreciated.

--
Patrick Glaser
Director of Research
McKinley Advisors


2233 Wisconsin Ave, NW, Suite 525 | Washington, DC 20007
27 N. Wacker Drive, Suite 444 | Chicago, IL 60606


Visit our blog 501(see).
Connect with us and other association executives on our LinkedIn Group.