Investigating how positive and negative feedback affects performance

 

 

 

 

Learning, Memory & Perception

 

 

Tutor: Aileen Proudlock

 

 

 

 

Dawn McManus

0057466

 

 

 

 

 

 

 

 

 

 

 

Introduction

 

 

The subject of reinforcement has been widely publicised and many studies and articles have probed the subject. Probably the most famous researcher in this area is Skinner. Although the title ‘Founding Father of Behaviour’ is attributed to B.F.Skinner, the roots of the approach go back to before the 19th century (Cardwell, Clark & Meldrum 1996: p521). Skinner himself acknowledged the influence of Thorndike in his work, especially his description of a ‘law of effect’. This law states that any behaviour that leads to a successful outcome will tend to be repeated in the future. (Thorndike 1911, cited in Cardwell, Clark & Meldrum 1996: p521).

 

Outcomes that are dependent on behaviour can be desirable or aversive. For instance, a positive stimulus is known as positive reinforcement and could promote the behaviour. However, a negative stimulus could be seen as a punishment and could serve to decrease the behaviour (Anderson 2000). The definitions of the different types of reinforcements and punishments we are concentrating on are simple. Positive reinforcement would be defined as a ‘pleasant stimulus that follows a desired behaviour’ and would increase the likelihood of the desired behaviour. Positive punishment would be defined as ‘Presentation of an unpleasant stimulus after an undesired behaviour occurs’ and could decrease the likelihood of the undesired behaviour (Atkinson et al 2000: p245). The main focus of this research project covered the aspects of positive reinforcement and positive punishment, or otherwise known as positive and negative feedback.

 

Many studies have looked into the idea of feedback and performance, and whether there is a relationship between the two. Berkowitz et al conducted an experiment on 25 eighteen to twenty eight year olds, who were asked to find the maximum height of an unseen hill by choosing a pair of co-ordinates. They received different levels of feedback on the height achieved. It was found that restricting the levels of feedback to the participant, significantly reduced performance (Berkowitz, Lewis & Drury 1983). This would suggest that feedback is an important tool for learning. Cohen & Herr found that out of 35 college teachers who participated in their study, many were rated higher by their students if they had administered interactive feedback, than those who were in the no feedback group (Cohen & Herr 1982). However feedback is not only a useful teaching tool. Philibin & Seidenstadt discovered from their time production experiment on 42 female undergraduates that partial feedback on 50% of the trials enhanced performance as effectively as 100% feedback did (Philibin & Seidenstadt 1983). This would indicate that feedback of any amount is helpful to performance. Matsui et al also looked at feedback and performance. They originally tested 87 female students who conducted a 10-minute arithmetic trial. After 5 minutes they received feedback on the number attempted. From this it was suggested that the feedback given induced a larger amount of effort from the participants. They replicated the study using 103 male students and this also found that feedback improved performance (Matsui Okada & Inoshita 1983). A similar study by Scriven & Glynn also indicates that feedback results in major gains to performance in given tasks (Scriven & Glynn 1983). It has to be said that although these studies indicate a positive reaction to feedback, they don’t distinguish whether the feedback is positive or negative. Pavett however did just this.

 

In her 1983 study on feedback and performance, it was hypothesised that the frequency of positive and negative feedback would make a significant contribution to the prediction of supervisory rated performance. The results indicated that amongst the 203 staff tested, positive rather than negative feedback served as an independent predictor of performance (Pavett 1983). An earlier study by Vitulli looked at not only positive and negative feedback, but also on no feedback at all. This study looked at the varying degrees of feedback on the number of correct guesses at a computerised ESP test. The computer program provided immediate feedback. Vitulli found that there was a significant mean score for each of the three feedback groups but a comparison test showed the largest mean difference between positive and negative feedback, with results indicating a close parallel between the effects of feedback and the number of correct guesses (Vitulli 1982). However a further Vitulli study on these three degrees of feedback on a computerised psi study, showed no effect (Vitulli 1983). Overall, with these studies in mind, it is suggested that feedback, whether positive or negative, does indeed have an impact on our learning performance.

 

 

 

Aim

As a result of the above literature, the aim of this study is to investigate the differential effects of positive and negative feedback.

 

 

Hypothesis

DV1 Reaction Time

H1: There will be a difference in reaction time between participants subjected to positive and negative feedback.

H0: There will be no difference in reaction time between participants subjected to positive and negative feedback.

 

DV2 Errors Made

H1: There will be a difference in errors made between participants subjected to positive and negative feedback.

H0: there will be no difference in errors made between participants subjected to positive and negative feedback.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Method

 

 

Design

This study was conducted using a semantic categorisation task with eight blocks of stimuli, either auditory or visual. The stimuli were randomised within each block for each participant. The order of the blocks was counter balanced, and the participant paused after each block. Auditory and Visual stimuli were matched on the basis of their word frequency, or how often we encounter the words in real life; and divided into large and small on the basis of the ratings by judges. There was a 1000 msec gap between the presentation of each stimulus. Each participant was given the same on screen instructions and heard the same auditory voice as a control measure. The study implements an independent measures design as each of the participants were subjected to only one level of the independent variable. In this case the independent variable was the feedback given, running on two levels: positive feedback and negative feedback. There are two dependent variables in this design. DV1 being the reaction time in each of eight blocks of tests after the feedback was given. DV2 is the errors made occurring in each of the eight blocks of tests, i.e. correct or incorrect answers.

 

Participants

The study was conducted using 78 naïve psychology undergraduates (N) of which 63 were female and 15 were male. The sample was opportunity as it was quick, easy and could be carried out in the Mac lab.

 

Materials

The materials used were a Macintosh computer with keyboard and monitor, a Superlab program, and suitable headphones.

 

Procedure

Each participant had to perform a semantic categorisation task. They read on screen instructions and responded to eight blocks of word stimuli by pressing relevant keys on the keyboard. The ‘a’ key was pressed when the word represented an object, which was smaller than the computer screen used, and the ‘l’ key was pressed when it was larger. There were 60 auditory stimuli and 60 visual stimuli. Half of all the stimuli represented objects that were larger, and the other half represented smaller ones. The stimuli were repeated between the eight blocks giving 360 experimental trials in total. Half of the participants received positive feedback: a tick for each correct answer, and half of them received negative feedback: a cross for each incorrect response. Their reaction times and errors were recorded. In two blocks they saw only visual stimuli; in another two blocks they heard only auditory stimuli, in the remaining four blocks they received a mixture of auditory and visual stimuli. The feedback was given 300 msec after each response. Participants were debriefed in accordance with BPS guidelines on the aims of the study during the next week’s seminar.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Results

 

 

Descriptive Statistics

A descriptive analysis of the participant’s group statistics was conducted. 78 participants took part and produced interval level data in the form of overall totals. The tests will be conducted using DV1 reaction time, DV2 errors and the IV feedback on two levels, positive and negative. The results are shown in figures 1 and 1a below:

 

 

 

Feedback

N

Mean

Standard. Deviation

Standard. Error Mean

Negative

35

7080.9199

3459.3275

584.7331
Positive

43

6284.7531

1286.3117

196.1607
Figure 1: Descriptive statistics of interval level data for all 78 participants and reaction time.

 

It can be seen from figure 1 above that there is a difference in means between those participants who were subjected to negative feedback, and those who were subjected to positive feedback; implying that participants receiving negative feedback reacted quicker. The standard deviations are also different with 3459.3 for negative and 1286.3 for positive showing that the negative group had more variation than the positive group.

 

 

 

 

Feedback

N

Mean

Standard. Deviation

Standard. Error Mean
Negative

35

32.0286

15.6759

2.6497

Positive

43

25.4651

11.7905

1.7980
Figure 1a: Descriptive statistics of interval level data for all 78 participants and error.

 

It can be seen from the figure 1a above that there is a difference in means between those participants who were subjected to negative feedback, and those who were subjected to positive feedback; again implying that participants receiving negative feedback created more errors. The standard deviations are also different with 15.6 for negative and 11.7 for positive showing that the negative group had slightly more variation than the positive group.

For further descriptive analysis, bar charts were created. Results are shown in figures 2 and 2a below:

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2: Bar chart showing differences in reaction time for the two levels of the IV

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2a: Bar chart showing differences in errors made for the two levels of the IV

 

The charts in figures 2 and 2a above clearly show the scale of differences in mean reaction time scores and errors made between participants who were subjected to negative and positive feedback. However it should be noted that the charts do not start at zero and could give misleading representations of the results due to the dramatic size differences.

 

Inferential Statistics

As the descriptive statistics are inconclusive, further analysis was needed. The data is interval level so parametric tests were used, in the form of  t-tests (see appendix 1).

The first test conducted was a Levene’s test. This is used to test for equality of variance. The results for DV1 and DV2 are shown in figures 3 and 3a below:

 

 

 

 

 

REACTION TIME

 

Levene's Test for Equality of Variances

 
F

Sig.
Equal variances assumed

4.140

.045
Equal variances not assumed

 

 
Figure 3: Levene’s test for equality of variance and reaction time.

 

It can be seen from figure 3 above that the significance is <0.05 (.045). This shows that the test does not have equality of variance despite being very close, and the results for the final t-test shown in figure 4 should be read from the bottom line where equal variances are not assumed.

 

 

 

 

 

ERRORS MADE

 

Levene's Test for Equality of Variances

 
F

Sig.
Equal variances assumed

1.787

.185
Equal variances not assumed

 

 
Figure 3a: Levene’s test for equality of variance and errors made.

 

It can be seen from figure 3a that the significance is >0.05 (.185). This shows that the test does have equality of variance, and the results for the final t-test in figure 4a should be read from the top line where equal variances are assumed. Results of the final t-tests for DV1 and DV2 are shown in figures 4 and 4a below:

 

 

 

 

Equal variances not assumed

 

REACTION TIME

 

 

 

t-test for Equality of Means

 

 
t

df

Sig. (2-tailed)
Equal variances assumed

1.397

76

.167
 

1.291

41.656

.204
Figure 4: Independent samples test on reaction time and feedback

 

By reading the bottom line in figure 4 above, where equal variances are not assumed, it can be seen that the significance level is >0.05 (.204) for a two tailed test. Therefore the standard form results are: t = 1.291; df = 41.656; p >0.05 (0.204).

 

 

 

 

Equal variances not assumed

 

ERRORS MADE

 

 

 

t-test for Equality of Means

 

 
t

df

Sig. (2-tailed)
Equal variances assumed

2.110

76

.038
 

2.050

61.897

.045
Figure 4a: Independent samples test on errors made and feedback

 

By reading the top line in figure 4a above, where equal variances are assumed, it can be seen that the significance level is <0.05 (.038) for a two tailed test. Therefore the standard form results are: t = 2.110; df = 76; p <0.05 (0.038).

 

 

 

 

Discussion

 

The aim of this study was to investigate the differential effects of positive and negative feedback. The results showed that there was no significance for effects of feedback on DV1 reaction time in relation to the experimental hypothesis, so H1 was rejected and H0 was accepted. For DV2 errors made, the results showed that there was significance in the results for feedback and errors made. Therefore in relation to the experimental hypothesis for DV2, H1 was accepted and H0 was rejected. The results for both tests were based on a two-tailed test.

 

These findings are true in relation to the literature that was used as the basis for this study. The studies shown focused on feedback and performance, with the majority, including studies by Matsui et al (1983) and Scriven & Glynn (1983), getting significant results. In the case of Vitulli (1982) there was found to be a strong link between feedback and correct guesses at a computer aided task. This is much in keeping with this particular study and specifically the hypothesis for DV2.

 

Despite the significance of this result for errors made, it is worth mentioning that errors could have occurred for other reasons over and above positive and negative feedback. For instance, the auditory voice was in a thick northern dialect, and some words could have been difficult to decipher resulting in errors. There is also an anticipatory issue; each participant would have been positioned in such a way that their fingers would have been hovering over the specific keys needed, in this case ‘a’ and ‘l’. Almost by default, if the participant had frequently pressed down the ‘a’ key, anticipation could have taken over and they could have involuntarily depressed the ‘l’ by mistake. Though these limitations are taken into consideration, the result was still significant for DV2 in such that feedback does indeed affect performance.

 

The fact that this study has continued in the vein of previous studies as seen in the introduction, can only hold welcome implications for the school of learning, and indeed future studies in the area. Along the same lines as Philibin & Seidenstadt’s study in 1983 which discovered that even 50% of feedback can enhance performance as much as 100% feedback can, further studies could investigate exactly how much feedback, either positive or negative can actually effect performance, and to what extent. If these techniques are regularly applied in both teaching, as shown in the Cohen & Herr study (1982), and learning as seen in the study by Berkowitz et al (1983) then the outlook for teaching professionals and students alike can be seen to be exciting at the very least.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

References

 

 

Anderson. J. R (2000) Learning and Memory: An Integrated Approach. 2nd Edition. USA: Wiley.

 

 

Atkinson. R. L; Atkinson. R. C; Smith. E. E; Benn. D. J & Nolen-Hoeksema. S (2000) Hilgard’s Introduction to Psychology. 13th Edition. USA: Harcourt Brace. p245.

 

 

Berkowitz. D; Lewis. R & Drury. C. G (1983) The Effect of Computer-Displayed Knowledge of Results on a Two-Variable Optimisation Task. Ergonomics. Vol. 26 (10) p 975-979 (Abstract).

 

 

Cardwell. M; Clark. L & Meldrum. C (1996) Psychology for A Level. London: Harper Collins. p 521.

 

 

Cohen. P. A & Herr. G (1982) Using an Interactive Feedback Procedure to Improve College Teaching. Teaching of Psychology. Vol. 9 (3) p 138-140. (Abstract).

 

Matsui. T Okada. A & Inoshita. O (1983) Mechanism of Feedback Affecting Task Performance. Organisational Behaviour & Human Decision Processes. Vol. 31 (1) p 114-122. (Abstract).

 

 

Philibin. T & Seidenstadt. R. M (1983) Feedback and Time Perception. Perceptual & Motor Skills. Vol. 57 (1) p 308. (Abstract).

 

 

Pavett. C. M (1983) Evaluation of the Impact of Feedback on Performance and Motivation. Human Relations. Vol. 36 (7) p 641-654. (Abstract).

 

 

Scriven. J & Glynn. T (1983) Performance Feedback on Written Tasks for Low Achieving Secondary students. New Zealand Journal of Educational Studies. Vol. 18 (2) p 134-145. (Abstract).

 

 

Vitulli. W. F (1982) Effects of Immediate Feedback on Computer Assisted Testing of ESP Performance. Psychological Reports. Vol. 51 (2) p 403-408. (Abstract).

 

 

Vitulli. W. F (1983) Immediate Feedback and Target Symbol Variation in Computer Assisted Psi Tests. Journal of Parapsychology. Vol. 47 (1) p 37-47. (Abstract).