Effects of Cognitive Load and Emotional Load on Memory Selectivity - An Individual Differences Perspective

MD & BJS

2022-06-26

In person data

Method

Participants

43 participants completed the study. We excluded 4 participants with low accuracy (< 70% correct) in the study phase and 2 with low accuracy in the test phase (hit - false alarm rate < 0). The final sample consisted of 10 women, 26 men, and one non-binary person. Mean age was 19 years (SD = 1).

check if four versions are counterbalanced across participants

For study phase:

expName n
V1_part1 9
V2_part1 9
V3_part1 10
V4_part1 9

For test phase:

expName n
V1_part2 9
V2_part2 9
V3_part2 10
V4_part2 9

Material

Procedure

  1. In the study phase, participants completed a task switching procedure was administered.

  2. As a filler task and to assess WMC, participants completed the letter-memory task.

  3. In the test phase, a surprise recognition memory test assessed participant’s memory for stimuli presented in the study phase

  4. Motivation measures: AATQ, BIS/BAS, Self-Control Scale (in random order)

Results

Study Phase

Descriptives

transition M_rt SD_rt SE_rt M_acc SD_acc SE_acc
repeat 1219 324 27 0.913 0.13 0.011
switch 2177 579 48 0.909 0.13 0.011

plot accuracy

plot response times

Inference

T-test on accuracy:

## 
##  Paired t-test
## 
## data:  acc by transition
## t = 0.47016, df = 36, p-value = 0.6411
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
##  -0.01306037  0.02094326
## sample estimates:
## mean of the differences 
##             0.003941441

T-test on reaction times:

## 
##  Paired t-test
## 
## data:  rt by transition
## t = -17.186, df = 36, p-value < 2.2e-16
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
##  -1070.7000  -844.6736
## sample estimates:
## mean of the differences 
##               -957.6868

The effects of task switching on performance (accuracy and reaction times) was replicated.

Test phase

We analyze only the Picture Recognition test because emotionality was induced by having negative and neutral words and we do not want to have a confound of material. Pictures were always neutral and they were counterbalanced across conditions and participants (words not).

Descriptives

Hits and False Alarms

Overall the hit rate was 0.535 (SE = 0.024) and the false alarms rate was 0.109 (SE = 0.012).

The first step is to replicate the task switching x attention interaction on recognition performance

Means per condition:

## `summarise()` has grouped output by 'attention'. You can override using the
## `.groups` argument.
attention transition m_acc sd_acc se_acc
target repeat 0.757 0.172 0.020
target switch 0.697 0.239 0.028
distractor repeat 0.301 0.173 0.020
distractor switch 0.387 0.199 0.023

plot recognition performance

ANOVA results:

## Warning: Collapsing data to cell means. *IF* the requested effects are a subset
## of the full design, you must use the "within_full" argument, else results may be
## inaccurate.
Effect DFn DFd SSn SSd F p p<.05 ges
(Intercept) 1 36 42.436 3.041 502.336 0.000 * 0.901
attention 1 36 5.424 0.920 212.348 0.000 * 0.538
transition 1 36 0.007 0.410 0.593 0.446 0.001
attention:transition 1 36 0.198 0.293 24.352 0.000 * 0.041

The interaction between attention and transition is critical here. The significant interaction replicates previous studies and means that task switching reduces memory selectivity.

This can also be seen in the main effect of transition on memory selectivity (Hits Targets - Hits Distractors).

Plot memory selectivity

This plot shows that memory selectivity is higher for repeat vs. switch trials. This is the task switching effect on memory selectivity.

Next, we introduce the factor Emotional Load to check the main effect on memory selectivity and a potential interaction with Cognitive Load (transition)

Please remember that we are looking only at memory for pictures (which were paired either with a emotional word or neutral word)

Before running the analysis for the full design (attention x transition x emotionality) let’s first check the interaction of emotional load with attention on recognition performance (hits)

Emotionality denotes if the word (which was presented over the picture) was negative or neutral

Emotional load seems to have little impact on memory

Descriptively, the pattern goes in the opposite direction compared to the pattern of the online data.

Let’s run the analysis for the full design. Again first for recognition performance and then for memory selectivity (which is the same just for better understanding of the meaning I show both)

2 x 2 x 2 ANOVA (attention x emotionality x transition)

Cognitive Load effect:

For Targets memory is better for repeat than switch

for Distractors memory is better for switch than repeat

Emotional Load effect:

no apparent differences

Effect DFn DFd SSn SSd F p p<.05 ges
(Intercept) 1 36 84.872 6.082 502.336 0.000 * 0.882
attention 1 36 10.848 1.839 212.348 0.000 * 0.488
emotionality 1 36 0.000 0.705 0.005 0.945 0.000
transition 1 36 0.014 0.820 0.593 0.446 0.001
attention:emotionality 1 36 0.003 0.601 0.202 0.655 0.000
attention:transition 1 36 0.396 0.586 24.352 0.000 * 0.034
emotionality:transition 1 36 0.006 0.341 0.634 0.431 0.001
attention:emotionality:transition 1 36 0.027 0.386 2.529 0.121 0.002

For remember responses:

Effect DFn DFd SSn SSd F p p<.05 ges
(Intercept) 1 36 34.634 7.768 160.501 0.000 * 0.695
attention 1 36 6.271 4.131 54.647 0.000 * 0.292
emotionality 1 36 0.015 0.464 1.139 0.293 0.001
transition 1 36 0.000 0.822 0.009 0.924 0.000
attention:emotionality 1 36 0.007 0.485 0.503 0.483 0.000
attention:transition 1 36 0.203 0.564 12.962 0.001 * 0.013
emotionality:transition 1 36 0.002 0.390 0.176 0.678 0.000
attention:emotionality:transition 1 36 0.061 0.608 3.612 0.065 0.004

For know responses:

Effect DFn DFd SSn SSd F p p<.05 ges
(Intercept) 1 36 11.073 5.220 76.369 0.000 * 0.528
attention 1 36 0.623 2.509 8.942 0.005 * 0.059
emotionality 1 36 0.012 0.367 1.218 0.277 0.001
transition 1 36 0.017 0.446 1.382 0.248 0.002
attention:emotionality 1 36 0.001 0.247 0.086 0.772 0.000
attention:transition 1 36 0.032 0.403 2.871 0.099 0.003
emotionality:transition 1 36 0.001 0.229 0.181 0.673 0.000
attention:emotionality:transition 1 36 0.007 0.487 0.501 0.484 0.001

Memory selectivity

Effect DFn DFd SSn SSd F p p<.05 ges
(Intercept) 1 36 21.697 3.678 212.348 0.000 * 0.761
emotionality 1 36 0.007 1.202 0.202 0.655 0.001
transition 1 36 0.793 1.172 24.352 0.000 * 0.104
emotionality:transition 1 36 0.054 0.772 2.529 0.121 0.008

There is no significant interaction between cognitive load and emotional load.

The next steps would be to introduce the individual differences variables.

Working memory capacity (WMC)

WMC was measured with the letter memory task.

ANCOVA

pes = partial eta squared ges = general eta squared

## Warning: Numerical variables NOT centered on 0 (i.e., likely bogus results):
## d.propCorrect
Effect df MSE F pes p.value
d.propCorrect 1, 35 0.10 0.19 .005 .668
transition 1, 35 0.03 1.16 .032 .290
d.propCorrect:transition 1, 35 0.03 5.19 * .129 .029
emotionality 1, 35 0.03 0.10 .003 .751
d.propCorrect:emotionality 1, 35 0.03 0.18 .005 .672
transition:emotionality 1, 35 0.02 0.03 <.001 .854
d.propCorrect:transition:emotionality 1, 35 0.02 0.03 <.001 .871

However, we need to center the variable on 0.

Effect df MSE F pes p.value
c.propCorrect 1, 35 0.10 0.19 .005 .668
transition 1, 35 0.03 27.18 *** .437 <.001
c.propCorrect:transition 1, 35 0.03 5.19 * .129 .029
emotionality 1, 35 0.03 0.20 .006 .659
c.propCorrect:emotionality 1, 35 0.03 0.18 .005 .672
transition:emotionality 1, 35 0.02 2.46 .066 .126
c.propCorrect:transition:emotionality 1, 35 0.02 0.03 <.001 .871

For recognition performance:

Effect df MSE F pes p.value
c.propCorrect 1, 35 0.17 0.34 .010 .566
attention 1, 35 0.05 207.55 *** .856 <.001
c.propCorrect:attention 1, 35 0.05 0.19 .005 .668
transition 1, 35 0.02 0.63 .018 .433
c.propCorrect:transition 1, 35 0.02 3.20 + .084 .082
emotionality 1, 35 0.02 0.00 <.001 .945
c.propCorrect:emotionality 1, 35 0.02 1.23 .034 .275
attention:transition 1, 35 0.01 27.18 *** .437 <.001
c.propCorrect:attention:transition 1, 35 0.01 5.19 * .129 .029
attention:emotionality 1, 35 0.02 0.20 .006 .659
c.propCorrect:attention:emotionality 1, 35 0.02 0.18 .005 .672
transition:emotionality 1, 35 0.01 0.62 .017 .437
c.propCorrect:transition:emotionality 1, 35 0.01 0.05 .001 .829
attention:transition:emotionality 1, 35 0.01 2.46 .066 .126
c.propCorrect:attention:transition:emotionality 1, 35 0.01 0.03 <.001 .871

3-way interaction (c.propCorrect x attention x transition) is not significant for remember and know

Centering with scale = FALSE (to indicate that we just want to subtract the mean)

Effect df MSE F pes p.value
c2.propCorrect 1, 35 0.10 0.19 .005 .668
transition 1, 35 0.03 27.18 *** .437 <.001
c2.propCorrect:transition 1, 35 0.03 5.19 * .129 .029
emotionality 1, 35 0.03 0.20 .006 .659
c2.propCorrect:emotionality 1, 35 0.03 0.18 .005 .672
transition:emotionality 1, 35 0.02 2.46 .066 .126
c2.propCorrect:transition:emotionality 1, 35 0.02 0.03 <.001 .871

In all three analyses the interaction between transition and WMC was significant. However, the main effect of transition changes…why?

What’s the problem with the WMC variable??!!

Check distribution of the mean proportion of correctly remembered letters…

Histogram raw data

It’s strongly skewed.

For proportions we can use the Arcsine Transformation

Histogram transformed data

Here are the descriptives for the raw and transformed data

N M SD Min Max Skewness Kurtosis Measure
37 0.83 0.19 0.33 1.00 -1.02 3.01 Proportion Correct
37 1.24 0.30 0.62 1.57 -0.32 1.89 Proportion Correct (arcsine transformed)

Even after the transformation the distribution does not look great (it seems more ordinal than continuous). It looks more like we have 3 groups.

For simplicity let’s use a median split for the analysis.

## [1] "Higher WMC" "Lower WMC"
OverallMedianSplit meanPropCorrect n
Higher WMC 0.9750000 20
Lower WMC 0.6666667 17

plot

2x2x2x2 ANOVA:

  • Attention (target vs. distractor)
  • Transition (switch vs. repeat)
  • WMC (higher vs. lower)
  • Emotionality (negative, neutral)
## Warning: Data is unbalanced (unequal N per group). Make sure you specified a
## well-considered value for the type argument to ezANOVA().
Effect DFn DFd SSn SSd F p p<.05 ges
1 (Intercept) 1 35 83.882 5.998 489.478 0.000 * 0.884
2 OverallMedianSplit 1 35 0.084 5.998 0.493 0.487 0.008
3 attention 1 35 10.610 1.740 213.443 0.000 * 0.490
5 emotionality 1 35 0.001 0.663 0.036 0.850 0.000
7 transition 1 35 0.010 0.786 0.453 0.505 0.001
4 OverallMedianSplit:attention 1 35 0.099 1.740 1.999 0.166 0.009
6 OverallMedianSplit:emotionality 1 35 0.042 0.663 2.217 0.145 0.004
8 OverallMedianSplit:transition 1 35 0.034 0.786 1.524 0.225 0.003
9 attention:emotionality 1 35 0.003 0.601 0.198 0.659 0.000
11 attention:transition 1 35 0.371 0.533 24.348 0.000 * 0.032
13 emotionality:transition 1 35 0.006 0.341 0.647 0.427 0.001
10 OverallMedianSplit:attention:emotionality 1 35 0.000 0.601 0.001 0.971 0.000
12 OverallMedianSplit:attention:transition 1 35 0.053 0.533 3.493 0.070 0.005
14 OverallMedianSplit:emotionality:transition 1 35 0.001 0.341 0.068 0.796 0.000
15 attention:emotionality:transition 1 35 0.025 0.377 2.274 0.141 0.002
16 OverallMedianSplit:attention:emotionality:transition 1 35 0.009 0.377 0.811 0.374 0.001

2x2x2 ANOVA on memory selectivity

## Warning: Data is unbalanced (unequal N per group). Make sure you specified a
## well-considered value for the type argument to ezANOVA().
Effect DFn DFd SSn SSd F p p<.05 ges
1 (Intercept) 1 35 21.220 3.480 213.443 0.000 * 0.765
2 OverallMedianSplit 1 35 0.199 3.480 1.999 0.166 0.030
3 emotionality 1 35 0.007 1.202 0.198 0.659 0.001
5 transition 1 35 0.742 1.066 24.348 0.000 * 0.102
4 OverallMedianSplit:emotionality 1 35 0.000 1.202 0.001 0.971 0.000
6 OverallMedianSplit:transition 1 35 0.106 1.066 3.493 0.070 0.016
7 emotionality:transition 1 35 0.049 0.755 2.274 0.141 0.007
8 OverallMedianSplit:emotionality:transition 1 35 0.017 0.755 0.811 0.374 0.003

With the median split analysis there is no sign. interaction between WMC and transition (result number 6).

Follow-up analyses

high WMC group

## Warning: You have removed one or more Ss from the analysis. Refactoring
## "subject" for ANOVA.
Effect DFn DFd SSn SSd F p p<.05 ges
(Intercept) 1 19 13.889 1.024 257.627 0.000 * 0.864
emotionality 1 19 0.003 0.473 0.126 0.727 0.001
transition 1 19 0.767 0.341 42.784 0.000 * 0.259
emotionality:transition 1 19 0.068 0.352 3.673 0.070 0.030

lower WMC group

## Warning: You have removed one or more Ss from the analysis. Refactoring
## "subject" for ANOVA.
Effect DFn DFd SSn SSd F p p<.05 ges
(Intercept) 1 16 8.007 2.455 52.175 0.000 * 0.650
emotionality 1 16 0.004 0.729 0.081 0.780 0.001
transition 1 16 0.132 0.725 2.920 0.107 0.030
emotionality:transition 1 16 0.004 0.403 0.146 0.707 0.001

The task switching effect on memory selectivity is only significant in higher WMC group This is in line with the online data. The number of participants per group is more balanced than in the online data.