Your Source for Learning
Technology, Strategy, and News
ARTICLES       RSS feed RSS feed

Research for Practitioners: Does Information Structure Affect Learning?

by Chris Atherton

May 30, 2013


by Chris Atherton

May 30, 2013

“The really interesting finding was that participants’ level of knowledge about the topic interacted significantly with the different learning conditions to affect how well they scored in the test. Crucially, linearly presented information appeared to be least suitable for non-knowledgeable participants, and most suitable for those who were knowledgeable about the subject.”

Is it better to present large amounts of information the same way to all learners, or should the presentation depend on the reader’s level of knowledge about the topic? The results of a study offer important insights about this question for eLearning design.

The study

“Influence of Text Structure and Prior Knowledge of the Learner on Reading Comprehension, Browsing, and Perceived Control.” Calisir, F. and Z. Gurel. Computers in Human Behavior, 19. 2003. (Editor’s Note: Unfortunately, the full text of this article is not readily available online, although at least one site will provide it for a fee.)

The question

How do learners at different levels of expertise engage with differently structured material? Specifically, should we give learners information in a long linear document (e.g., a single-page article) or in a hierarchical format, where they can see the structure and choose links to different parts of the document? Also, is this different for novice learners vs. learners who already have knowledge of the subject?

The method

Participants in this study read a piece of text around 5,000 words long. Of the 30 participants, half were knowledgeable about the subject, having taken a course on it, and half were not. Participants accessed the text on a computer in one of three randomly assigned conditions:

  1. The first group read the text as a linear document, like the kind you might produce in Word.
  2. The second group were given the text as a hierarchically-structured hypertext document: Imagine the Wikipedia page for your favorite TV show, which has several “child” pages, one for each series of the show; each series page in turn has a number of “child” pages, one per episode, etc. In this study, the hierarchical structure went six layers deep.
  3. The last group read the text as a “mixed” hypertext document, similar to the hierarchical group, but with several additional “network-style” links between pages—in the Wikipedia example, extra links like this might connect two episodes starring the same guest actor or connect a series page and an episode page whose titles share a common theme.

So in other words, while the text was the same in all three conditions, participants were able to navigate it in three very different ways.

All participants were given 40 minutes to read the text, during which time the computer tracked the links clicked by those in the hierarchical and mixed groups. A short test was then used to measure participants’ reading comprehension.

The results

First, the unsurprising results: Participants who were knowledgeable about the material performed significantly better in the reading comprehension test than their less knowledgeable peers. This in itself isn’t too surprising, since when you already know something about a topic, it’s much easier to understand and retain new information about it.

Furthermore, there was no significant difference between the test scores of knowledgeable and non-knowledgeable participants across the hierarchical and mixed conditions—this makes sense when you consider that the information in both conditions was hyperlinked with a dominant hierarchical structure, and that participants in both groups clicked about the same number of links while reading.

The really interesting finding was that participants’ level of knowledge about the topic interacted significantly with the different learning conditions to affect how well they scored in the test. Crucially, linearly-presented information appeared to be least suitable for non-knowledgeable participants and most suitable for those who were knowledgeable about the subject. See Figure 1.

Graph showing average reading comprehension scores for knowledgeable and non-knowledgeable participants in the  linear, hierarchical and mixed presentation groups. Linear, Non-knowledgeable: 13.40. Linear, Knowledgeable: 20.80. Hierarchical, Non-knowledgeable: 18.60. Hierarchical, Knowledgeable: 20.20. Mixed, Non-knowledgeable: 17.60. Mixed, Knowledgeable: 17.60.)

Figure 1: Average reading comprehension scores for knowledgeable and non-knowledgeable participants in the linear, hierarchical, and mixed presentation groups

Implications for eLearning design

  1. Consider that long linear documents may be better suited to expert learners. As experts, these learners may already have internalized many of the material’s implicit structural hierarchies—reiterating such structural information might be unnecessarily repetitive, and could result in something like the “expertise reversal effect” (Kalyuga et al).
  2. Help non-expert learners compensate for their lack of mental models of the domain by providing structural information, such as hierarchical relations. Note that a strictly “network-like” hyperlinked environment can disorient beginners (see Mohageg), so it’s a good idea to preserve some recognizable hierarchy.

Additional references:

Kalyuga, S., P. Ayres, P. Chandler, and J. Sweller. “The Expertise Reversal Effect.” Educational Psychologist, 38 (1). 2003.

Mohageg, M. F. “The Influence of Hypertext Linking Structures on the Efficiency of Information Retrieval.” Human Factors, 34. 1992.

Topics Covered

Appreciate this!
Google Plusone Twitter LinkedIn Pinterest Facebook Email Print

Login or subscribe to comment

This article certainly gave me pause in thinking about its implications for structuring content. I would certainly be interested in hearing further discussion about the results (which might just mean finding the full article, or hearing more from the author and other readers)...

1. I would imagine that pure words/content of any form would be tough for a novice, as opposed to some sort of structured activity (or even a directed reading/research activity, with a set of targeted questions or other focus). From this summary, it was a little hard to tell how well anyone actually performed on the comprehension test, and how predictive the test would be of future behavior. I guess my (cynical) assumption would be that there would be lots of other formats could have a more positive impact on performance, but in the absence of other options, these results might suggest one format over another?

2. While expert audiences might be able to absorb information from a linear source, from a practical perspective, the more expert an audience becomes, the more likely it feels that they won't want a 5,000-word document as a resource. :) They want the 1/2-page executive, bulleted version, scannable in 40 seconds as opposed to 40 minutes. I'd be curious if the study included any data on the interaction with the content, e.g. if the more expert group truly read all of the words in any of the formats! I could envision that all of the experts would perform about the same on the test (which they did), but then maybe a few of the linear group, who like that sort of thing, actually read it and picked up enough information to improve their performance on the test.

Quite curious,

-- Michael Wolfe
I would benefit from hearing a robust explanation of linear vs hierarchical structure. Does anyone have suggestions where I might find a deeper explanation of the differences? My initial assumption is this is a bottom-up vs top-down approach, linear being bottom-up and hierarchical being top-down. Top down would start with concepts and ideas approached at a 5000 foot level conceptually, then a closer view at the 1000 foot level to build up knowledge, then an even closer view of individual constructs that make up the topic.
Thanks both for your comments and interest!

Michael, about your first point, sorry, I'm not quite clear what you're asking. The figures reported here are the reading comprehension scores. I'm sure there are many other factors involved in reading comprehension, but the authors of this paper chiefly sought to manipulate these aspects. They did also (which I didn't mention partly for reasons of brevity, but also because it wasn't significant) look at the "perceived control" experienced by research participants — that is, the extent to which they felt in control over how and in what order they read the document. However, they found that the three types of document didn't seem to affect perceived control significantly. What kind of "future behaviour" were you thinking of?

Your second point is a really interesting one. The authors didn't find a significant effect of the number of links clicked (in the hierarchical and mixed conditions) on reading comprehension. Obviously there was no equivalent measurement for the linear condition. In an ideal world, eye-tracking would have been nice, or at least data on scrolling behaviour! It's possible that experts are better able to scan material quickly, of course.

Cheryllyne, the distinction the authors describe between linear and hierarchical is that linear is literally one set of information after another, whereas hierarchical allows users to go down and back up the trees of parent-child content at will. The authors sadly don't specify the order, in the linear document, of general (5000ft) vs specific (say, 1000 or 100ft) information — which is a shame actually! It would be really interesting to know the order in which that information was presented in the linear condition — not least because good writing (IMO) often seems to begin with the general and then drill down to the specific.
Personally, the lesson I took away from this (without, I hope, reading anything more into the data than is actually there) is to simply avoid lengthy exposition where the learners are novices. As a general strategy, it appears that what the study described as a hierarchical document (structured so the readers can drill down at their discretion and as they feel they need) would best serve everyone, regardless of level of expertise. The non-knowledgeable can drink their fill. The knowledgeable can drink of such as they require to complete their knowledge -- which may mean drinking very little or even none of the information offered. All learners can "satisfice" themselves with a hierarchical document, according to their need and inclination. Or so it seems to me.
It’s great to see real experimental rather than speculative data presented in the field of L&D. In particular, the importance of prior knowledge is fascinating. It’s almost as if less knowledgeable participants need to build a complex web of understanding whereas those familiar with the topic will be helped by a linear approach driving a railroad across their existing web of understanding to get to their destination.
Related Articles