Effects of knowledge map characteristics on information processing

CONTEMPORARY EDUCATIONAL PSYCHOLOGY 17, 136155 (1992)

Effects of Knowledge Map Characteristics on Information Processing

DOUGLAS A. WIEGMANN, DONALD F. DANSEREAU, EDWARD C. MCCAGG, KIRSTEN L. REWEY, AND URVASHI PITRE

Texas Christian University

The majority of research on knowledge maps has focused on demonstrating their effectiveness in educational settings. However, very little work has attempted to examine the parameters of map construction that mediate their impact on performance. The present set of experiments represents an attempt to remedy this situation by examining the impact of three parameters of map construction on knowledge acquisition. Experiment 1 investigated spatial configuration, Experiment 2 investigated map format, and Experiment 3 investigated link structure. Results of these examinations revealed that the way the maps were constructed influenced both the encoding and the retrieval of the information in the maps and that these effects were mediated by the users’ spatial and verbal abilities. 0 1992 Academic

Press, Inc.

Knowledge maps are two-dimensional spatial arrays that represent information in the form of a node-link-node diagram. See Fig. 1 for an example of a typical knowledge map. Over the last several years, a great deal of research has been conducted to examine knowledge maps as mul- tipurpose educational tools (e.g., as text supplements and substitutes, Hall, 1988; lecture aids, Holley & Dansereau, 1984; prewriting tools, McCagg & Dansereau, 1989; note-taking strategies, Novak & Gowin, 1984; and props during cooperative interactions, Rewey, Dansereau, Skaggs, Hall, & Pitre, 1989). In general, the results of this body of research suggest that under certain conditions, knowledge maps can be a more performance-effective alternative than traditional text.

To date, however, the majority of research on knowledge mapping has been primarily demonstrational. There have been virtually no attempts to examine the critical parameters of map construction and presentation, This situation mirrors the pattern that has occurred with graphical displays in general (Kosslyn, 1989). In short, little is known about how to construct effective graphical displays or what makes one graphical dis-

This research was supported in part by National Science Foundation Grant MDR8751369, Donald F. Dansereau, Principal Investigator. Any opinions, findings, conclusions, or rec- ommendations expressed in this material are the authors’ and do not necessarily reflect the views of the National Science Foundation. Address correspondence and reprint requests to Douglas A. Wiegmann, Department of Psychology, Texas Christian University, Ft. Worth, TX 76129.

136

0361-476X/92 $3.00 Copyright 0 1992 by Academic Press, Inc. All rights of reproduction in any form reserved.

KNOWLEDGE MAPS, SEMANTIC MAPS 137

FIG. 1. Example of a typical knowledge map and link types. T, type; C, characteristic; P, part; R, results in, L, leads to; Ex, example.

play more effective than another (Larkin & Simon, 1987). The present collection of experiments represents a combination of attempts to remedy this situation.

Three experiments are reported that independently examined the impact of variations in one of three structural properties common to most map displays (i.e., configuration, format, and links; Kosslyn 1989). Since the experiments are a combination of projects, they vary slightly in terms of their stimulus materials and dependent measures. However, the experiments are similar in terms of their design and methodology. When con- sidered together, they provide related information about the impact that variations in map characteristics have on the process of learning from knowledge maps.

KNOWLEDGE MAP CHARACTERISTICS

Spatial ConJiguration

There are several ways in which knowledge maps can be configured spatially, and these configurations vary across different systems of mapping (Lambiotte, Dansereau, Cross, & Reynolds, 1989). For example, maps can be configured in the form of a web. Web configurations consist of a topic node in the center of the map surrounded by supporting nodes. The result is a map that looks like a spider’s web. Other techniques include hierarchical configurations, for example, placing the topic node

138 WIEGMANN ET AL.

on the top center of the map and then placing subordinate concepts be- neath it. This sequence continues down the page with the bottom nodes being the most subordinate concepts in the domain. In general, however, the effectiveness of particular configurations in improving performance is still unknown.

Kosslyn (1989) and others (e.g., Lambiotte et al., 1990) have suggested that maps configured in a manner congruent with the processing priorities of the perceptual system (e.g., configurations that use gestalt organizational principles of symmetry, proximity, and good continuation) should be more performance-effective than maps that do not incorporate these principles. To test this assumption, students in Experiment 1 were presented with a knowledge map on human biology. The map was either configured using gestalt principles of organization (i.e., symmetry, proximity, and good continuation) or configured in a web-like manner (i.e., topic node surrounded by supporting nodes) and did not use gestalt organization principles. See Figs. 2 and 3 for schematic representations of these two types of maps.

Map Format

Knowledge maps can be formatted as a single, large map (i.e., whole map; see Fig. 2) or as a smaller set of interlocked, cross-referenced maps (i.e., stacked maps; see Fig. 4; Lambiotte et al., 1989). Whole maps may have advantages in that they provide an overall view of the domain and multiple processing routes through the information. However, whole maps may have problems associated with size and complexity that po- tentially reduce their usefulness for instructional purposes (Lambiotte et al., 1989). Stacked maps, on the other hand, sequentially present seg-

FIG. 2. A schematic representation of a gestalt map presented in a whole-map format.


FIG. 3. A schematic representation of the web-like configuration used in Experiment 1.

ments of the domain in smaller chunks. Consequently, in addition to being more manageable, stacked maps allow readers to generate their own overall organization of the domain. Nonetheless, since the information is presented sequentially, stacked maps may have problems associated with a limited number of processing routes through the domain.

Although the different formats may have their own unique advantages and disadvantages, the majority of mapping systems tend to utilize the

Map 1

Map 2

Map 3

I

Map 4

FIG. 4. A schematic representation of a stacked map.

140 WIEGMANN ET AL.

whole-map format (Lambiotte et al., 1989). The effectiveness of the two formats, however, has yet to be compared directly. To address this issue, students in Experiment 2 were presented information on human biology in the form of a whole map (i.e., a single, large map) or in the form of a stacked map (i.e., a set of smaller, interlocked, cross-referenced maps).

The Links

Links in a knowledge map represent information about the relationship between concepts. Fundamentally, links consist of a single line that con- nects two nodes (Novak 8z Gowin, 1984). Additional information can be added to the links by (a) using arrowheads to specify the directionality of the relationships (i.e., superordinate-subordinate relationships), (b) using words or abbreviations to label (i.e., name) the relationships, and (c) using differently shaped lines to classify the types of relationships (e.g., solid vs barbed lines to represent static vs dynamic relationships, respectively; see Fig. 1; Lambiotte et al., 1989).

In order to examine the effects of variations in links on performance, students in Experiment 3 were presented with a knowledge map on human biology. The links in the map either consisted of single, unembellished lines or they contained the additional information described above (i.e., arrowheads, labels, and shape). Although most mapping systems utilize some combination of these link components to designate relationships, the two extremes (i.e., lines vs embellished links) were compared in order to increase the likelihood of detecting any potential impact that variations in linking may have on performance.

ASSESSING PERFORMANCE

A second common objective of the present set of experiments was to examine the impact that variations in map properties have on the stages of information processing (i.e., encoding and retrieval). Consequently, each experiment used two measures that varied in the degree to which they provided cuing for the retrieval of the information. A test with few cues (e.g., a recall test) requires that the answers be actively retrieved from long-term memory. A test that provides many cues (e.g., a recognition test) requires that the answers be matched with the information encoded during studying. Therefore, the less cuing on the measure, the higher the demands on retrieval processes; the greater the cuing on the measure, the more sensitive to encoding processes.

In each experiment, the test that provided the most cuing was a multiple-choice test that simply required the recognition of correct answers. The tests that placed a greater demand on retrieval processes (i.e., provided less retrieval cues) were a fill-in-the-blank test (Experiment l), a short-answer test (Experiment 2), and a free-recall test (Experiment 3). If,

KNOWLEDGE MAPS. SEMANTIC MAPS 141

in each experiment, the effect size (i.e., the magnitude of the treatment; w*) is larger on the test that places a higher demand on retrieval processes, it would imply that variations in the map(s) had some impact on retrieval. If the effect size is the same or less, it would seem likely that the map variations impacted only on encoding.

FINAL CONSIDERATIONS

A further issue related to the investigation of map properties stems from equivocal results emerging from research that examined the roles of spatial and verbal abilities in map processing (e.g., Dees & Dansereau, 1990; Rewey, Dansereau, Skaggs, Dees, & Pitre, 1990; Skaggs, 1988). One potential reason for this equivocality is that each of the studies used different maps with different structural properties. It is possible that these properties have different effects on students who differ in abilities (Kosslyn, 1989). To assess this possibility, a third common objective of the present set of experiments was to examine the relationship between spatial and verbal abilities and the impact of variations in map parameters on learning from knowledge maps.

Two reliable and commonly used tests of spatial and verbal abilities are the Group Embedded Figures Test (GEFT; Oltman, Raskin, & Witkin, 1971), and the Delta Reading Vocabulary Test (Delta; Deignan, 1973), respectively (see Dansereau, 1978; Hall, 1988; Rewey et al., 1990). Con- sequently, these two tests were used to assess students’ spatial and verbal abilities in the present experiments.

The GEFT (Oltman et al., 1971) requires participants to locate a series of simple geometric shapes embedded within more complex geometric patterns. High scores on this test reflect an individual’s ability to disem- bed configurations within a spatial display (Hall, 1988). The Delta (Deignan, 1973) consists of single target words followed by five options. The participant’s task is to indicate the option word that is synonymous with the target. Scores on this test have been shown to be correlated moderately (r = .60) with other, more elaborate measures of verbal ability (i.e., scores on the verbal section of the Scholastic Aptitude Test; Dansereau, 1978).

EXPERIMENT 1

The purpose of this experiment was to examine the effects of variations in map configuration (i.e., gestalt vs web) on performance of students differing in spatial and verbal abilities.

Method Subjects. Thirty-seven students from Texas Christian University completed both phases

of this experiment. All participants were enrolled in an undergraduate psychology course and received class credit for their participation.

142 WIEGMANN ET AL.

Practice materials. The practice materials consisted of an example knowledge map con- taining embellished links (i.e., the links had labels, arrowheads, and shape), displayed on a single 11 X 17-m. sheet of paper. Half of the map was configured using gestalt principles of organization, the other half was configured in the form of a web and did not incorporate gestalt principles.

Study materials. The study materials consisted of two isomorphic maps (consisting of 80 nodes and links) of the human autonomic nervous system (ANS). The information included a description of the structures, processes, and functions of the sympathetic and parasympathetic nervous systems. Each map was displayed in a whole-map format (i.e., single, large map) on a 24 x 24-in. sheet of paper and contained embellished links. The maps differed only in terms of their spatial configuration.

Specifically, two configurations were used (i.e., gestalt vs web; see Figs. 2 and 3). The gestalt map was configured using the principles of (a) symmetry (e.g., parallel information between the sympathetic and parasympathetic systems was represented using symmetrical node arrangements), (b) proximity (e.g., information about the effects of the sympathetic nervous system were clustered together in relation to other information), and (c) good continuation (e.g., information about the way the parasympathetic system processes information was presented in a sequence of nodes and links that continued in a straight line from the first node in the process to the last).

In the web version, which contained the same nodes and links as the gestalt map, the information was represented by placing the topic node (i.e., autonomic nervous system) in the center of the page and supporting information around it. The principles of symmetry, proximity, and good continuation were not used to organize the configuration of the map.’

Testing materials. Two tests were used in order to assess students’ memories of the ANS information. To avoid cuing answers across measures, the tests were administered in order of increasing retrieval cues (i.e., decreasing demands on retrieval processes). The first test, till-in-the-blank, consisted of 30 sentences about the ANS that had key terms deleted; participants were required to recall the correct term for each blank. The second test, multiple-choice, was identical to the till-in-the-blank test; however, on this test, participants were asked to recognize the correct answer by choosing one of four alternatives.

Individual difference measures. The Group Embedded Figures Test (Oltman et al., 1971) consists of 25 items and takes 18 min to administer. The Delta Reading Vocabulary Test (Deignan, 1973) consists of 45 items and takes 10 min to administer.

Procedure. The experiment consisted of two, l-h sessions. During the first session, participants were introduced to the knowledge mapping technique (15 min). During the introduction, an experimenter described the different aspects of the map (i.e., the nodes and the links) and demonstrated how node-link-node connections can be translated into simple declarative sentences. Participants were then paired with a partner and allowed 5 min to practice reading an example map to each other.

Following the practice period, participants were administered the Delta (10 mitt). Upon completion of the Delta, participants were assigned randomly to one of the two map groups: web-map (n = 17) or gestalt-map (n = 20). Participants were informed that they would be

i In order to determine the external validity of the two map configurations, an additional two groups of college students were presented with either the gestalt map (n = 10) or the web map (n = 10). Following a 4-min study period, students completed three, eight-point Likert scale questions aimed at assessing their initial impressions of the map. A series of t tests (p < .05) indicated that students exposed to the gestalt map reported that the map was more organized and that it was easier to identify the similarities and differences between the subareas of information.


tested thoroughly over the information during the next session. They were then allowed 20 min to study the ANS information presented in the map.

Participants began the second session (1 day later) by completing the fill-in-the-blank test over the ANS material. They were given 20 min to complete this test and then allowed 15 min to complete the multiple-choice test. Finally, students were administered the GEFT (18 min).

Results

For all analyses reported in this section, an (Y level of .05 is implied when an effect is reported as significant.

Scoring. The fill-in-the-blank test, the multiple-choice test, the GEFT, and the Delta were scored using predetermined scoring keys. A participant’s score on each of these measures was computed by summing the number of correct responses on each test.

Preliminary analyses. Pearson product-moment correlations were computed to examine the relationship between scores on the two individual difference measures (i.e., GEFT and Delta) and scores on the two memory tests (i.e., fill-in-the-blank and multiple-choice). The results of these analyses revealed that scores on the GEFT and the Delta correlated positively and consistently with performance of students across both map groups (correlations ranged from r = .24 to r = .32).

Simple regression analyses indicated that, across groups, participants’ scores on the GEFT and Delta were linearly related to their performance on both memory tests. Comparisons of the within-group regression lines using an approach recommended by Keppel (1982, p, 505) revealed no significant differences between the slopes of the lines. Therefore, it was decided that scores on the GEFT and Delta would be utilized as covari- ates in subsequent analyses.

Primary analyses. Table 1 presents the mean percentages and standard deviations of each test (i.e., fill-in-the-blank and multiple-choice) as a function of map group (gestalt vs web). Note that group means have been

TABLE 1 EXPERIMENT 1: MEANS AND STANDARD DEVIATIONS FOR PERCENTAGE CORRECT ON

TESTS AS A FUNCTION OF MAP GROUP

Map group Fill-in-the-blank Multiple-choice

Gestalt map (n = 20) M 73.56 80.30 SD 19.61 19.48

Web map (n = 17) M 54.12 66.27 SD 18.28 17.41

Note. Group means have been adjusted for differences in spatial and verbal abilities.

144 WIEGMANN ET AL.

adjusted for differences in spatial and verbal abilities. An examination of Table 1 reveals that the adjusted mean scores of students in the gestalt- map group were ordered higher than the adjusted mean scores of students in the web-map group on both tests.

In order to determine the reliability of these observations, a multivariate analysis of covariance (MANCOVA) was performed on the fill-in-the- blank and multiple-choice test scores. Map group (gestalt vs web) was the between-group factor. GEFT and Delta scores were utilized as covari- ates. A MANCOVA was used in lieu of a repeated-measures approach because the MANCOVA creates a linear combination of dependent vari- ables that produces a more reliable test of group differences (Hair, Ander- son, & Tatham, 1987).

The results of the analysis revealed a significant multivariate effect for map group, F(2,32) = 5.48. Univariate post hoc analyses indicated that students in the gestalt-map group reliably outperformed students in the web-map group on both the fill-in-the-blank, F(1,33) = 11.25, MSe = 272.74, w* = .23, and the multiple-choice tests, F(1,33) = 5.76, MSe = 277.14, w* = .ll.

Discussion

The results of the present study suggest that the way in which a knowledge map is configured influences performance. Students who studied the knowledge map configured using gestalt principles of organization outperformed students who studied a map configured in a web-like manner that did not incorporate these principles.

Supposedly, a map configured using gestalt organizational principles is congruent with certain processing priorities of the readers’ spatial/ perceptual systems (Kosslyn, 1989). This congruency may have facilitated the acquisition of the information by providing an organizational scaffold and allowing better navigation through the display (Lambiotte et al., 1989).

In addition, indices of effect magnitude revealed that map configuration had a greater impact on students’ fill-in-the-blank test performance (w* = .23) than on their multiple-choice test performance (w* = . II). Since the fill-in-the-blank test supposedly placed higher demands on retrieval processes than the multiple-choice test, it appears that the gestalt map may have facilitated retrieval as well as encoding.

Finally, the results of the present experiment also suggest that individual differences play a mediational role in processing differentially contig- ured knowledge maps. Students’ spatial and verbal abilities (i.e., their scores on the GEFT and Delta) were related positively to their performance on both the multiple-choice and the till-in-the-blank tests.


EXPERIMENT 2

The purpose of this experiment was to examine the effects of variations in map format (i.e., whole map vs stacked map) on performance of students differing in spatial and verbal abilities.

Method Subjects. Thirty-four students from Texas Christian University completed both phases of

this experiment. All participants were enrolled in an undergraduate psychology course and received class credit for their participation.

Pracrice materials. The practice materials consisted of two example knowledge maps that contained embellished links (i.e., the links had labels, arrowheads, and shape) and were configured using gestalt principles of organization. One map was formatted as a whole map on an 11 x 17-in. sheet of paper; the other was formatted as a set of stacked maps displayed on 8.5 x 11-m. sheets.

Study materials. The study materials consisted of isomorphic maps (consisting of 80 nodes and links) on the ANS. The maps contained embellished links and were configured using the gestalt principles of organization described in Experiment 1. The maps differed only in terms of their format.

The maps were formatted either as a single, large map (i.e., whole map) on a 42 X 16-in. sheet of paper or as a set of six smaller, interlocked, cross-referenced maps (i.e., stacked maps; see Fig. 4) displayed on 8.5 X 1 l-in. sheets. The dimensions of the whole map differed in two ways from those of the whole map used in Experiment 1. First, it was presented on a smaller sheet of paper. Second, it was not arranged in a hierarchical diamond shape as shown in Fig. 2. Rather, the subsections of the information were placed at the same level on the page and lined up sequentially from left to right.

Testing materials. Two tests were used to assess students’ memory of the ANS information. To avoid cuing answers across measures, the tests were administered in the order of increasing retrieval cues. The first test, short-answer, consisted of five questions that required students to recall specific information about the structure, processes, and functions of the sympathetic and parasympathetic nervous systems. The second test, multiple-choice, was similar to the test used in Experiment 1 and consisted of 25 questions.

Individual difference measures. Similar to Experiment 1, the Group Embedded Figures Test (Oltman et al., 1971) and the Delta Reading Vocabulary Test (Deignan, 1973) were used as indicators of students’ spatial and verbal abilities, respectively.

Procedure. The procedure was similar to Experiment 1. During the first session, participants were introduced to the knowledge mapping technique and then practiced reading the example maps to each other (20 min). Following the practice period, participants completed the Delta Reading Vocabulary Test (10 min).

Upon completion of the Delta, participants were assigned randomly to one of the two map groups: whole-map (n = 18) or stacked-map (n = 16). Participants were informed that they would be thoroughly tested over the information during the next session. They were then allowed 20 min to study the ANS information presented in the map(s).

Participants began the second session (1 day later) by completing the short-answer (20 min) and multiple-choice (10 min) tests over the ANS information. Next, the GEFT was administered (18 min). Students were than thanked and dismissed.

Results

In this section, an (Y of 0.5 is implied when an effect is reported as significant.

146 WIEGMANN ET AL.

Scoring and reliability. Scoring of the short-answer test was based on a propositional scoring technique developed by Skaggs (1988). A scoring key was constructed for the test by dividing the material on the ANS into idea units. Each idea unit contained one fact expressed in the form of a simple declarative sentence.

An experienced rater matched the idea units in the scoring key with the idea units contained in participants’ short-answer tests. For each idea unit that appeared in both the key and a participant’s test, the participant received a score from one to four depending on the accuracy of the match. The sum of these individual scores constituted a participant’s overall score for the test. Reliability of the scoring system was established by having a second experienced rater score 15 randomly selected tests. A reliability coefficient of r = .98 was achieved.

The multiple-choice test, the GEFT, and the Delta were scored using predetermined scoring keys. A participant’s score on each measure was computed by summing the number of correct responses on each test.

Preliminary analyses. A series of correlations were computed to examine the relationship between scores on the Delta and GEFT and scores on the short-answer and multiple-choice tests. The results of these analyses indicated that scores on the Delta correlated positively with performance across map groups (correlations were r = .23 and r = .38).

Simple regression analyses indicated that, across groups, participants’ performance on the Delta was linearly related to their performance on both the short-answer and the multiple-choice tests. Comparisons of the within-group regression lines revealed no significant differences between the slopes of the lines. Therefore, it was decided that Delta scores could be used as a covariate in subsequent analyses.

The results of the analyses using GEFT scores revealed nonhomoge- neous correlations across dependent measures as a function of map group. Correlations were r = - .06 and r = - .15 in the whole-map group and r = .46 and r = .68 in the stacked-map group for the multiple-choice and short-answer tests, respectively. Simple regression analyses indicated that GEFT scores significantly predicted the performance of students in the stacked-map group but did not predict the performance of participants in the whole-map group.

In order to further examine the relationship between GEFT scores and test performance as a function of map group, scores on the GEFT were used to form a second between-groups variable in subsequent analyses. A median split was performed on GEFT scores, dividing students into high (n = 18) and low (n = 16) spatial ability groups.

A median split was used in lieu of a regression approach in order to examine participants’ test performance as a function of map group and spatial ability while taking into account variance attributable to verbal


ability. A reexamination of the use of Delta scores as a covariate using both map group and spatial ability as factors indicated that the homoge- neity of regression slopes assumption had not been violated. Therefore it was decided that test performance would be examined in a two-way factorial design using map group and spatial ability as between-groups factors and Delta scores as a covariate.

Primary analyses. Table 2 presents the mean percentages and standard deviations of each test (i.e., short-answer and multiple-choice) as a function of map group (whole vs stacked) and spatial ability (high vs low). Note that group means have been adjusted for differences in verbal ability. An examination of Table 2 reveals that adjusted mean scores of low spatial ability students in the whole-map group were ordered higher than adjusted mean scores of low spatial ability students in the stacked-map group. On the other hand, adjusted mean scores of high spatial ability students in the stacked-map group were ordered higher than adjusted mean scores of high spatial ability students in the whole-map group.

In order to examine the reliability of group differences, a 2 (whole vs stacked map) x 2 (high vs low spatial ability) MANCOVA was performed on the short-answer and multiple-choice test scores. Delta scores were utilized as the covariate (see Experiment 1 for rationale concerning the use of a MANCOVA in lieu of a repeated measures approach).

The results of this analysis revealed a significant two-way multivariate interaction between map group and spatial ability, F(2,28) = 3.98. No


TESTS AS A FUNCTION OF MAP GROUP AND SPATIAL ABILITY

Map group Short-answer Multiple-choice

Whole map Low spatial (n = 8)

M SD

54.10 67.07 34.46 20.17

High spatial (n = 10) M SD

32.18 56.76 27.51 23.48

Stacked map (n = 8) Low spatial

M SD

34.24 60.82 31.94 23.61

High spatial (n = 8) M SD

66.41 77.35 13.01 17.69

Note. Group means have been adjusted for differences in verbal ability.

148 WIEGMANN ET AL.

other multivariate effects were significant. Univariate analyses revealed that the interaction between map group and spatial ability was significant using short-answer test scores as the dependent measure, F( 1,29) = 8.19, MSe = 732.36, w2 = .18, and marginally significant using multiple-choice test scores, F(1,29) = 3.92, p = .057, MSe = 386.03, w2 = .08.

Analyses of simple effects indicated that, although the mean scores of low spatial ability students in the whole-map group were ordered higher than low spatial ability students in the stacked-map group, these differences were not significant. Comparisons at the level of high spatial ability, however, revealed that high spatial ability students in the stacked-map group reliably outperformed high spatial ability students in the whole-map group on both the short-answer, F(1,29), = 6.93, w2 = .27, and the multiple-choice tests, F(1,29) = 4.95, w2 = .19. No other comparisons were made.

Discussion

The results of the present study suggest that different methods of for- matting map displays (i.e., whole vs stacked) differentially influence performance depending on spatial ability. The stacked-map format facilitated the performance of high spatial ability students and tended to hinder the performance of low spatial ability students. In contrast, the whole-map format facilitated the performance of low spatial ability students and tended to hinder the performance of high spatial ability students.

Stacked maps present information in segmented chunks and allow users to form their own overall organization of the domain (Lambiotte et al., 1989). Presumably, stacked maps allowed high spatial ability students to capitalize on their proficiency in processing and arranging spatial information. The stacked maps may have hindered low spatial ability students due to their lack of proficiency in these skills.

It is unclear why the whole map compensated low spatial ability students while apparently hindering high spatial ability students. This finding seems incongruent with the findings of the previous experiment which suggested that high spatial ability students can benefit from studying a whole map. However, as mentioned earlier, the dimensions of the whole map used in this experiment differed in two ways from the whole map used in Experiment 1. First, the map was presented on a smaller sheet of paper and second, it was not arranged in the hierarchical diamond shape as shown in Fig. 2. Rather, the subsections of the information were placed at the same level on the page and lined up sequentially from left to right.

Possibly, the sequential nature of the information in the whole map used in this experiment facilitated the use of traditional text reading strategies (i.e., reading from left to right, and from top to bottom). Conse-


quently, the whole map may have compensated low spatial ability students for their lack of proficiency in processing spatial material by allowing them to read the map like they would a text. On the other hand, the sequential nature of the map may have hindered high spatial ability students from capitalizing on their proficiency in processing spatial information.

The results of the present study also suggest that map format and spatial ability may interact to influence both the encoding and the retrieval of the information presented in the map(s). Indices of effect magnitude indicated that the interaction between map format and spatial ability had a greater impact on students’ short-answer test performance (w2 = .18) than on their multiple-choice test performance (w2 = .08). Since the short-answer test supposedly placed higher demands on retrieval processes, it appears that map format and spatial ability may interact to mediate both the encoding and the retrieval stages. These effects however, seemed to occur mostly at the level of high spatial ability.

Finally, the results of the present study support the finding in Experi- ment 1 that verbal ability plays a role in processing map displays. Stu- dents’ scores on the Delta were related positively to their performance on all test measures.

EXPERIMENT 3

The purpose of this experiment was to examine the effects of variations in the representation of relationships (i.e., lines vs embellished links) on performance of students differing in spatial and verbal abilities.

Method

Subjects. Thirty-one students from Texas Christian University completed both phases of this experiment. All participants were enrolled in an undergraduate psychology course and received class credit for their participation.

Practice materials. The training materials consisted of an example knowledge map con- taining embellished links. The map was formatted in a stacked-map format and configured using gestalt principles of organization.

Study materials. The study materials were modified versions of the stacked maps used in Experiment 2. The maps contained 80 nodes and links and were displayed on six, 8.5 x

1 l-in. sheets of paper. Both sets of maps were configured using gestalt principles of organization (e.g., symmetry and good continuation; see Experiment 1). The sets of maps differed only in terms of their link structure. The links either contained labels, arrowheads, and contigural information (i.e., barbed vs solid lines; see Fig. 1) or they were unembellished, straight lines.

Testing materials. Two tests were used to assess students’ memories of the ANS information. In order to avoid cuing answers across tests, the tests were administered in order of increasing retrieval cues. The first test, free-recall, required students to recall everything they could remember about the ANS material. The second test, multiple-choice, was similar to those used in the previous experiments and consisted of 15 items.

Individual difference measures. Similar to the preceding experiments, the Group Embed-

150 WIEGMANN ET AL.

ded Figures Test (Oltman et al., 1971) and the Delta Reading Vocabulary Test (Deignan, 1973) were used as indicators of students’ spatial and verbal abilities, respectively.

Procedure. The procedure was similar to the previous experiments. During the first session, participants were introduced to the knowledge mapping technique and then practiced reading an example map to each other. Following the practice period, participants completed the Delta Reading Vocabulary Test (10 min).

Upon completion of the Delta, participants were assigned randomly to one of the two map groups: lines (n = 15) or embellished links (n = 16). Participants were informed that they would be tested thoroughly over the information during the next session; they were then allowed 20 min to study the ANS information presented in the map.

Participants began the second session (1 day later) by completing the short-answer (20 min) and multiple-choice tests over the ANS information. Then the GEFT was administered (18 min), and students were thanked and dismissed.

Results

For all analyses reported in this section, an cx level of .05 is implied when an effect is reported as significant.

Scoring and reliability. Scores and reliabilities for the free-recall test were computed using the propositional scoring technique described in Experiment 2. A reliability coefficient of r = .95 was achieved.

The multiple-choice test, the GEFT, and the Delta were scored using predetermined scoring keys. A participant’s score on each of these measures was computed by summing the number of correct responses on each test.

Preliminary analyses. Pearson product-moment correlations were computed to examine the relationship between scores on the GEFT and the Delta and scores on the free-recall and multiple-choice tests. The results of these analyses revealed that scores on the GEFT correlated moderately with performance of students across both map groups. Cor- relations were r = .30 and r = .32.

Simple regression analyses indicated that, across groups, participants’ performance on the GEFT was a linear predictor of their performance on both the free-recall and the multiple-choice tests. Comparisons of the within-group regression lines revealed no differences between the slopes of the lines. Therefore, it was decided that GEFT scores could be utilized as a covariate in subsequent analyses.

The results of the analyses using Delta scores revealed nonhomoge- neous correlations across dependent measures as a function of map group. Correlations were r = - .08 and r = - .12 in the lines group, and r = .45 and r = .60 in the embellished-links group on the multiple-choice and free-recall tests, respectively. Simple regression analyses indicated that Delta scores significantly predicted the performance of students in the embellished-links group but did not predict the performance of participants in the lines group.

In order to further examine the relationship between participants’ per-


formance on the Delta and their performance on the two memory tests as a function of map group, Delta scores were used to form a second between-groups variable in subsequent analyses. A median split was performed on Delta scores, dividing students into high (n = 15) and low (n = 16) verbal ability groups.

A median split was used in lieu of a regression approach in order to examine participants’ test performance as a function of map group and verbal ability while taking into account variance attributable to spatial ability. A reexamination of the use of GEFT scores as a covariate using both map group and verbal ability as factors indicated that the homoge- neity of regression slopes assumption was not violated. Therefore it was decided that test scores would be examined in a two-way factorial design employing map group and verbal ability as between-groups factors and GEFT scores as a covariate.

Primary analyses. Table 3 presents the mean percentages and standard deviations for each test (i.e., free-recall and multiple-choice) as a function of map group (lines vs embellished links) and verbal ability (high vs low). Note that group means have been adjusted for differences in spatial ability. An examination of the table reveals that adjusted mean scores of low verbal ability students in the lines group were ordered higher than adjusted means scores of low verbal ability students in the embellished-links group. On the other hand, adjusted mean scores of high verbal ability students in the embellished-links group were ordered higher than adjusted mean scores of high verbal ability students in the lines-group.


TESTS AS A FUNCTION OF MAP GROUP AND VERBAL ABILITY

Map group Free-recall Multiple-choice

Lines Low verbal (n = 7)

M SD

High verbal (n = 8) M SD

Links Low verbal (n = 9)

M SD

High verbal (n = 7) M SD

25.54 49.71 16.75 26.86

13.07 35.66 8.10 22.09

12.75 41.71 7.89 12.37

28.86 64.36 16.% 29.42

Note. Group means have been adjusted for differences in spatial ability.

152 WIEGMANN ET AL.

In order to determine the reliability of these observations, a 2 (links vs lines) x 2 (high vs low verbal ability) MANCOVA was performed on the free-recall and multiple-choice test scores. GEFT scores were utilized as the covariate (see Experiment 1 for rationale concerning the choice of a MANCOVA).

The results of this analysis revealed a significant two-way multivariate interaction between map group and verbal ability, F(2,25) = 5.45. No other multivariate effects were significant. Univariate analyses revealed that the interaction between map group and verbal ability was significant for both free-recall, F(1,26) = 10.81, MSe = 136.66, +v2 = .26, and multiple-choice tests, F(1,26) = 5.17, MSe = 479.94, w2 = .12.

Analyses of simple effects indicated that mean scores of low verbal ability students in the lines group were significantly higher than mean scores of low verbal ability students in the embellished-links group on the free-recall test, F(1,26) = 4.70, w2 = .16, but not on the multiple-choice test. In addition, high verbal ability students in the embellished-links group reliably outperformed high verbal ability students in the lines group on both free-recall, F(1,26) = 6.87, w2 = .28, and multiple-choice tests, F(1,26) = 6.43, w2 = .34. No other comparisons were made.

Discussion

The results of the present study suggest that different methods of rep- resenting relationships (i.e., embellished links vs unembellished lines) differentially influence performance depending on verbal ability. Link embellishments facilitated the performance of high verbal ability students, but tended to hinder the performance of low verbal ability students. On the other hand, the absence of these link embellishments facilitated the performance of low verbal ability students and tended to hinder the performance of high verbal ability students.

Link embellishments provided additional information about the specific nature of the relationships between nodes (Lambiotte er al., 1989). The presence of these embellishments presumably allowed high verbal ability participants to capitalize on their proficiency in processing semantic material, while hindering low verbal ability participants due to their lack of proficiency. The absence of these link embellishments presumably re- duced the demand on semantic processing, allowing the opportunity for low verbal ability students to focus their attention on processing key ideas within the nodes. In contrast, high verbal ability students may have sub- merged themselves in the task of deciphering the unembellished links, consequently spending less time on processing the main ideas.

The results of the present study also suggest that link embellishments and verbal ability may interact to influence both the encoding and the


retrieval of the information presented in maps. Indices of effect magnitude indicated that the interaction between link structure and verbal ability had a greater impact on students’ free-recall test performance (w2 = .26) than on their multiple-choice test performance (w2 = .12). However, this impact appeared to be more a function of low verbal ability. Effect sizes at the level of high verbal ability tended to lean more toward an encoding effect (i.e., w2 = .28 vs w2 = .34, for the recall and multiple- choice tests, respectively).

Finally, the results of the present experiment support the findings in Experiment 2 that high spatial ability students benefit from studying a stacked-map format. Students’ scores on the GEFT were related positively to their performance on ail measures.

GENERAL DISCUSSION

The results of the present series of experiments suggest that the way in which knowledge maps are constructed influences their impact on performance. In addition, variations of certain map components (i.e., format and links) tended to influence performance differently depending on the particular abilities of the user (i.e., spatial and verbal abilities). Finally, the impact of the various structural properties and their interaction with ability tended to influence both encoding and retrieval processes.

Although these findings are of interest in their own right, further research is needed to determine whether variations in knowledge map characteristics impact performance when domains other than the autonomic nervous system are used. In addition, further research is needed to examine whether the results found in the present set of experiments occur when knowledge maps are varied along more than one structural dimen- sion. Nonetheless, it seems clear that the structural properties of map displays can be identified and that their impact on performance can be studied systematically.

The results of the present study also have broader implications for a preliminary model of spatial/verbal processing recently proposed by Lam- biotte et al. (1989). The model is based on theories oriented toward the processing of statistical graphs (Pinker, 1985) and geographical maps (Kulhavy, Lee, & Caterino, 1985), as well as more general theories of spatial cognition (e.g., Kosslyn, 1987).

According to the model, the presentation of a graphical display acti- vates both the spatial (i.e., perceptual and imagic) and the verbal (i.e., syntactic and propositional) processing systems of the user. These systems, in turn, contribute either independently or interactively to intluence processing during the encoding and retrieval stages.

During encoding, the user’s spatial system uses the salient configural properties of the display to locate major concepts and guides the verbal

154 WIEGMANN ET AL.

processing system in extracting surrounding detailed, propositional information. During retrieval, an individual’s spatial processing system (with its retention of locational and configural information) guides the retrieval of the major concepts; the verbal system then provides supporting propositional information.

The results of the present study suggest that variations in map configuration, format, and link structure can mediate these stages of spatial/ verbal processing. It appears that future work may need to be focused on elaborating the encoding and retrieval stages of spatial/verbal processing to account for these findings. Such elaborations may involve developing a framework of how maps influence the construction of display-based knowledge representations (i.e., micro- and macrostructures) and specu- lations about how the map properties influence the interaction between the spatial and the verbal systems in moving from display-based information to situational models (van Dijk & Kintsch, 1983).

REFERENCES DANSEREAU, D. F. (1978). The development of a learning strategies curriculum. In H. F.

O’Neil, Jr. (Ed.), Learning strategies (pp. l-29). New York: Academic Press. DEES, S. M., & DANSEREAU, D. F. (1990, April). The strategic processing of knowledge

maps as text supplements and substitutes. Paper presented at the Annual Meeting of the American Educational Research Association, Boston, MA.

DEIGNAN, G. M. (1973) The Delta Reading Vocabulary Test. Air Force Human Resources Laboratory, Lowry Air Force Base, CO.

HAIR, J. F., ANDERSON, R. E., & TATHAM, R. L. (1987). Multivariate data analysis (2nd ed.). New York: Macmillan.

HALL, R. (1988). Knowledge maps and the presentation of related information domains. Unpublished doctoral dissertation, Texas Christian University, Fort Worth, TX.

HOLLEY, C. D., & DANSEREAU, D. F. (1984). Networking: the technique and the empirical evidence. In C. D. Holley & D. F. Dansereau (Eds.), Spatial learning strategies: Tech- niques, applications, and related issues (pp. 81-108). New York: Academic Press.

KEPPEL, G. (1982). Design and analysis: A researchers handbook. New Jersey: Prentice- Hall.

KOSSLYN, S. M. (1987). Seeing and imagining in the cerebral hemispheres: A computational approach. Psychological Review, 94, 148-175.

KOSSLYN, S. M. (1989). Understanding charts and graphs. Applied Cognitive Psychology, 3, 185-226.

KULHAVY, R. W., LEE, J. B., & CATERINO, L. C. (1985). Conjoint retention of maps and related discourse. Contemporary Educational Psychology, 10, 28-37.

LAMBIOTTE, J. B., DANSEREAU, D. F., CROSS, D. R., & REYNOLDS, S. B. (1989). Multire- lational semantic maps. Educational Psychology Review, 1, 331-367.

LARKIN, J. H., & SIMON, S. A. (1987). Why a diagram is (sometimes) worth ten thousand words. Cognitive Science, 11, 65-99.

MCCAGG, E. C., & DANSEREAU, D. F. (1989, April). Knowledge mapping as a pre- communication technique. Paper presented at the Annual Meeting of the Southwestern Psychological Association, Houston, TX.

NOVAK, J. D., & GOWIN, D. B. (1984). Learning how to learn. New York: Cambridge Univ. Press.


OLTMAN, P. K., RASKIN, E., & WITKIN, H. A. (1971). The Group Embedded Figures Test. Palo Alto, CA: Consulting Psychologists Press.

PINKER, S. (1985). Visual cognition: An introduction. In S. Pinker (Ed.), Visual cognition (pp. l-63). Cambridge, MA: MIT Press.

REWEY, K. L., DANSEREAU, D. F., SKAGGS, L. P., HALL, R. H., & PITRE, U. (1989). Effects of knowledge maps and scripted cooperation on the recall of technical material. Journal of Educational Psychology, 81, 604409.

REWEY, K. L., DANSEREAU, D. F., SKAGGS, L. P., DEES, S. M., & PITRE, U. (1990, April). Scripted cooperation and knowledge map supplements: Effects on recall of biological and statistical information. Journal of Experimental Education, in press.

SKAGGS, L. P. (1988). The effects of knowledge maps and pictures on the acquisition of scientific information. Unpublished doctoral dissertation, Texas Christian University, Fort Worth, TX.

VAN DIJK, T., & KINTSCH, W. (1983). Strategies of discourse comprehension. New York: Academic Press.

Effects of knowledge map characteristics on information processing

Documents

Transcript of Effects of knowledge map characteristics on information processing