Feedback x color . . . 1

Feedback on error does not impact the association weight of contiguously active traces

Roy B. Clariana, Assistant Professor, Instructional Systems (email: )

The Pennsylvania State University, Great Valley School of Graduate Professional Studies, 30 E Swedesford Road, Malvern, PA 19355; Office phone: 610-648-3253;

Presented at the 6th International Conference on Cognitive and Neural Systems (ICCNS), May 30, 2002, at Boston College, Boston, MA.

Abstract

Does instructional feedback affect memory of extra-item context variables? This investigation examined explicit posttest memory of lesson color schemes. Lesson treatments used either constructed response or multiple-choice questions with immediate feedback. Posttest memory of lesson screen color was no better than chance guessing (20%) when the lesson required a recall response, but color was better remembered when the lesson required a recognition response. Explicit posttest memory of lesson color context was unrelated to the amount of lesson feedback provided.

Among a number of connectionist learning rules, the delta rule (Shanks, 1995; Widrow & Hoff, 1960) is one of the simplest and most common that includes the effects of feedback on learning. The delta rule describes the change in association weight, termed Dw, between an input unit and an output unit at each learning trial, as Dwio = a ain (to - aout), where a is the learning rate parameter, ain is the activation level of input units, to is the desired response (the t refers to "teacher", in this case to is item feedback), and aout is the activation level of the output units (Shanks, 1995). In instructional terms, learning is an increase in association, that is, an increase in Dwio between the stimulus (ain) and the correct response (aout), with a relative decrease in association, that is, a decrease in Dwio for incorrect responses. Clariana, Wagner, and Rorher-Murphy (2000) have shown that learning gains due to the effects of feedback can be accurately predicted by the delta rule (Clariana, 1999, 2000). They used lesson item difficulty as a measure of initial unit output activation before feedback and posttest item difficulty as a measure of final unit output activation. Their experimental data showed a difference for verbatim and inferential lesson questions (see Figure 1).


Figure 1. Observed delta rule increase in score from lesson to posttest.

This begs an important question, does instructional feedback have a specific effect only on to-be-remembered (TBR) content, or does it also have a general effect on other co-occurring activated units (i.e., context)? For example, acquiring a phobia through the association of a neutral context variable with some adverse circumstance. If co-occurring contextual information is affected by instructional feedback, this could support explanations of the effects of contextual variables on memory.

Lesson color scheme was selected as a non-content related context variable in this investigation. In the memory literature, background screen color is classified as a context variable that is extra-item (not a direct part of the TBR content), local (i.e., on the same screen as the TBR content, as opposed to global, which refers to the larger environment), and nonverbal (Mori & Graf, 1996). A contextual element like color may or may not be directly related to the TBR content, but may end up mixed in with the learner’s memories of that content anyway (Weiss & Margolius, 1954).

Context is important in learning if a context-dependency or memory context effect (MCE) occurs. For example, changing global context between lesson and test, such as underwater and on land (Godden & Baddeley, 1975, 1980), has been shown to negatively affect recall test performance but not recognition test performance. Recognition and recall response modes are regularly associated with MCEs in the memory literature. Further, though feedback is an important instructional component, no research has considered the effects of feedback on memory of color context. The purpose of this investigation is to examine the encoding and retrieval effects of instructional feedback on color as a local, extra-item, and non-verbal context variable under recognition and recall response modes. In this investigation, lesson item difficulty directly relates to amount of feedback received during the lesson. A correlation analysis will address the question of whether posttest memory of color context relates to the amount of lesson feedback received. Specifically, does lesson feedback only affect TBR content, or does feedback have a general effect on all contiguously activated traces (i.e., color)?

Method

Participants

The participants (n = 35) in this investigation were graduate student volunteers from the Corporate Training Design and Development program at a northeastern university. Participants used self-selected pseudonyms during the lesson task and posttest. 78% of the participants are full time employees of local corporations and businesses, and 20% are school teachers; most were female (75%) and all but one are in their late-twenties.

Materials

The lesson content consisted of the definitions of common instructional design terms used by Clariana and Lee (2001). The 36 terms were arranged in the lesson in a logical order based on the order of their first occurrence in the introductory instructional design textbook (Dick & Carey, 1996), and then were divided into five groups each assigned a color. These groups include: Overview (orange, hexadecimal RGB value #FF6531) – items 1 through 9, Instructional Analysis (yellow, #FFFF00) – items 10 through 16, Learner Analysis (blue, #0030FF) – items 17 through 22, Objectives and Tests (green, #00FF00) – items 23 through 27, and Development and Evaluation (purple, #9C0063) – items 28 through 36. These color hues are not normally associated with these terms or groups (such as using red to indicate a concept such as “stop”).

The color associated with each group was designed to appear as merely decorative and the participants were not informed that the color groupings reflected related groups of concepts. The color was displayed on the screen as a solid column 1.4 inches wide along the left margin, similar to the design of many web pages. The remainder of the screen (the right column) was white and about 5.8 inches wide. This white area served as the main text display area, though some cuing information was displayed in the colored area. The colored left column format was only used in the lesson, the posttest used uniformly white screens without color (see Figure 2).


Figure 2. Sample MC lesson screen with color bar.

Two computer-delivered lessons, one multiple-choice (MC) and the other constructed response (CR), were developed in Authorware 4.0. The computer program randomly assigned participants to either the CR or the MC lesson. The MC lesson provided the 36 terms in multiple-choice format with one term per screen (see Figure 2). First, the definition was provided along with four possible responses. After a correct response, the learner was told "Right", and the correct alternative was highlighted in blue with the message “Here is the answer”. On error, the learner was told, "No, look above", as the TBR content was highlighted in blue with the message “Here is the answer”.

The CR lessons provided the 36 terms in constructed-response format, again with one term per screen. First, the definition was provided along with a blank text box. The screen stated, “Type in the correct term and then press the Enter key.” The feedback stated "Right" for a correct response and the correct alternative was highlighted in blue with the message “Here is the answer”. On error, the feedback stated, "No, try again" and a list of 40 instructional design terms would appear at the bottom of the screen. The student was allowed one more try with the list available. If the second-try response was correct on the second try, the feedback stated, "Right”, and the correct alternative was highlighted in blue with the message “Here is the answer”. If the second-try response was incorrect, the feedback stated, "Here is the correct response", and the correct alternative was displayed highlighted in blue.

Procedure and Design

During regular class time, the instructor explained the purpose of this activity. Participants were asked to do their best during the lesson and posttest. Participants completed the lessons and posttest in the computer lab during regular class time. With both CR and MC lesson formats, students advanced to the next screen at their own pace by clicking a Continue button. Students completed their assigned lesson lessons and then the immediate posttest in about 25 minutes.

Posttest

All participants received the same 4-alternative multiple-choice posttest recognition posttest. Participants were instructed to click on the correct term (see Figure 3). Next, five color bars appeared beneath the item with the text “OK, now pick a color that goes with the question.” As soon as the participant clicked on one of the colors, the program advanced to the next term, repeating the answer and color selection process. Cronbach alpha reliability for TBR content was 0.75 for the recognition posttest, while alpha for recognition color memory was 0.67.


Figure 3. Sample MC posttest screen with color choices.

Results

Lesson and posttest means and standard deviations are shown in Table 1. The CR lesson and MC lesson obtained equivalent posttest scores for memory of TBR content. However, posttest memory of the lesson color scheme was significantly greater for the MC lesson treatment compared to the CR lesson treatment (t-test probability = 0.02). McDaniel and Mason (1985) state, "initial recognition and recall tests do not have similar effects on existing memory representations" (p.381). They suggest that recall tasks (like CR here) elaborate existing memory traces, providing richer and more meaningful integration with existing semantic memory, while in contrast, recognition tasks (like MC here) strengthen existing traces including non-semantic and contextual information such as color.

Table 1. Means and standard deviations (in parentheses) for each treatment.

Lesson / Posttest
Treatments / TBR
content / TBR
content / Color memory
CR lesson
(n=16) / 25.8
(6.3) / 31.4
(3.8) / 8.8
(3.6)
MC lesson
(n=19) / 29.6
(3.8) / 31.4
(2.6) / 12.6
(4.8)

What effect does feedback on error have on contextual information? Does immediate feedback on error also increase the association weight of color context information, or are feedback effects specific to the TBR content only? Specifically, at the instance of feedback processing, both TBR content and color context are likely to both be active, and so feedback may have the general effect of increasing the association weight of all active traces (such as learning a phobia by the incidental pairing of the object and an unpleasant experience), or feedback may have a specific effect on just the traces that are directly related to the feedback, in this case, the TBR content. To answer this question, color memory posttest item difficulty values from the posttest were correlated with their corresponding TBR content lesson item difficulty for each treatment. A negative correlation would indicate that feedback (after an error) increases the color context association weight, a positive correlation would suggest that feedback decreases the color context association weight, and no correlation would indicate that there is little relationship between amount of lesson feedback and posttest memory of color context information.

Posttest memory of TBR content (see solid lines in Figure 4) followed the classic form of the delta rule, with the greatest gain occurring with the greatest amount of feedback. However, the amount of lesson feedback had little relationship to explicit posttest memory of lesson color (see dashed lines in Figure 4), all of the correlations were small and non-significant (r = 0.06 and r = -0.03). Apparently, feedback has a specific effect on explicit memory of TBR content rather than a general effect on other active memory traces.


Figure 4. TBR Content (solid) and Color (dashed) posttest memory for CR and MC lesson treatments as a function of the amount of lesson feedback (on error).

Discussion

This investigation considered the effects of feedback on posttest memory of an extra-item context variable (lesson color svhemes) and also on memory of TBR content. Posttest memory of screen color was no better than chance guessing when the lesson required a recall response (CR), but was better remembered when the lesson required a recognition response (MC). Further, explicit posttest memory of lesson color scheme was unrelated to the amount of lesson feedback provided.

Color plays a critical and central role in nature, for example the recognition of both safe and dangerous food. Our neural systems may be especially tuned to color, since those individuals with better color processing were slightly more likely to survive and pass that ability on to the collective gene pool. Thus color may be a very special kind of context variable and so the effects for color should not be automatically applied to other context variables.

In this investigation, color hue may have influenced lesson and posttest performance. Specifically, the best posttest memory of TBR content occurred with the green color hue and the worst with the yellow hue, while posttest memory of color context showed the opposite order. Apparently, color hue is not a unitary variable. For example, McConnohie (1999) used color-coded power point slides (white, blue, or green background) to present sets of randomly selected letters and numbers, and then assessed memory using free recall (i.e., no color cue at testing). The observed trends suggest a slight superiority for white over blue and green, especially on delayed memory of slide content. Similarly, Petrich and Chiesi (1976) reported significantly greater free recall for red over green background color in a paired-associate task. Mori and Graf (1996) in a list learning study using words displayed with a color band on power point like slides observed a superiority for red over green for explicit recognition performance and for green over red for implicit (perceptual) recognition performance. If color hue differentially affects cognitive processing as suggested here, this will substantially complicate interpretation of previous color context research and confounds color context studies that have inadequately controlled color hue (which is most studies including the present one).

References

Clariana, R. B. (1999). Differential memory effects for immediate and delayed feedback: A delta rule explanation of feedback timing effects. A Presentation at the Annual Convention of the Association for Educational Communications and Technology, February, 1999 in Houston, TX. (ERIC Document Reproduction Center ED 430 550).