Introduction
Stress s a pervasive psychological experience, that experts found has significant effects on cognitive funtioning, memory in particular (Czekalla et al., 2021). Research has found that stress impair the encoding and retrival of information, although the nature of this is widely debated (Czekalla et al., 2021). Other research suggests that there is a linear decline in memory when stress increases. While other research points to complex, non-linear patterns. Agruing that moderate stress can enhance some aspects of recall before performance deteriates under stressful conditions (Schnabe et al., 2022). In foresic contexts, it is important to note that stress shapes memory. A witness or participants often expreience stressful circumstances and the ability to accurately recall details can have significant consequences (Hellhammer, 2026). The purpose is to evaluate the effectiveness of four level stress manipulation. By examining its impact on memory, performance across different tasks (Joel et al., 2023). The aim of the assignment is to determine if stress manipulation produces reliable differences in affective and behavioural measures. Lastly, the aim of the assignment is to see if there is a linear or non-linear trajectory.
Methods and Results
The data analysis will involve exploring data with visualisations, a one way anovo’s on manipulation check of variables (negative affect scores and subjective stress). A regression modeling of memory outcomes with a polynomial contrast applied to test for potential non-linear relationships between memory and stress levels. The lineup accuracy will examine logistic regression.
#variables Descriptives
vars <- c("stress_lvl", "negative_affect_score", "lab_stuff_score", "sss_conf_sum",
"sss_socio_sum","sss_cons_sum")
desc <- describe(s2_complete_data [, vars])
desc[, c("mean", "sd", "median", "range")]
## mean sd median range
## stress_lvl* 2.42 1.11 2 3
## negative_affect_score 14.88 4.62 13 23
## lab_stuff_score 13.09 1.94 13 11
## sss_conf_sum 10.23 2.33 10 11
## sss_socio_sum 16.40 4.13 17 18
## sss_cons_sum 20.38 6.72 20 32
The descriptive statistics were ran on all continuous variables. Stress levels is low to moderate (Mean = 2.42, SD = 1.11 and the median = 2. Negative affect scores averaged = 14.88, SD = 4.63 and median = 13. Lab staff scores also average = 13.09, SD = 1.94 and median = 1. The the sss subscales scores were high for conscientiounsness sum (Mean = 20.08, SD = 6.72 and median = 20. Sociability (Mean = 16.46,SD = 4.13 and median = 17. While confidence showed a (Mean = 10.23, SD = 2.33 and median 10).
Manipulation Check
#Negative Affect Sum
anova_neg_aff<-aov(negative_affect_score ~ stress_lvl, data = s2_complete_data)
summary (anova_neg_aff)
## Df Sum Sq Mean Sq F value Pr(>F)
## stress_lvl 3 86 28.55 1.345 0.262
## Residuals 162 3438 21.22
TukeyHSD(anova_neg_aff)
## Tukey multiple comparisons of means
## 95% family-wise confidence level
##
## Fit: aov(formula = negative_affect_score ~ stress_lvl, data = s2_complete_data)
##
## $stress_lvl
## diff lwr upr p adj
## 2-1 0.57015810 -1.9515556 3.0918718 0.9359105
## 3-1 -1.25478469 -3.9030932 1.3935238 0.6087690
## 4-1 0.53468900 -2.1136195 3.1829975 0.9531671
## 3-2 -1.82494279 -4.4464358 0.7965503 0.2737283
## 4-2 -0.03546911 -2.6569622 2.5860239 0.9999844
## 4-3 1.78947368 -0.9540138 4.5329612 0.3305977
To assess whether the stress manipulation was effective, a one-way anova was conducted to with stress levels (4 level)as the independent variable and negative affect score as the dependent a variable. Descriptive showing a Mean = 14.88. sd = 4.62 and median = 13.00 and a homogenity of variance can be assumed. The anova test showed no significance F(3,162)=1.35 and p= .262. This shows that negative affect did not significantly differ across all four stress levels. The sum of squares between groups was 86 with an observation of 3438,which also showed a total variance of 2.4% by stress.
#Confidence Sum
anova_sss_conf<-aov(sss_conf_sum ~ stress_lvl, data = s2_complete_data)
summary(anova_sss_conf)
## Df Sum Sq Mean Sq F value Pr(>F)
## stress_lvl 3 53.2 17.742 3.421 0.0188 *
## Residuals 158 819.3 5.186
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 4 observations deleted due to missingness
TukeyHSD (anova_sss_conf)
## Tukey multiple comparisons of means
## 95% family-wise confidence level
##
## Fit: aov(formula = sss_conf_sum ~ stress_lvl, data = s2_complete_data)
##
## $stress_lvl
## diff lwr upr p adj
## 2-1 1.0731225 -0.17373678 2.319982 0.1183931
## 3-1 1.2607656 -0.04868844 2.570220 0.0637617
## 4-1 1.4759358 0.12578292 2.826089 0.0261157
## 3-2 0.1876430 -1.10855210 1.483838 0.9818345
## 4-2 0.4028133 -0.93448432 1.740111 0.8624731
## 4-3 0.2151703 -1.18067231 1.611013 0.9782228
A one-way anova test showed a significant effect of stress level as sss confidence sum F(3,138)= 3.42, p= .019. The Tukey post hoc showed that participants in stress level 4 reported a significantly higher confidence scores than those of stress level 1 (Mean diff= 1.48, p= .026 and 95% CI [0.13, 2.83]). The difference between level 3 and 1 was marginal (Mean diff= 1.26, p= .064). No other pairwise comparisons were significant (all p> .11).
#Conscientiousness
anova_sss_cons<-aov(sss_cons_sum ~ stress_lvl, data = s2_complete_data)
summary(anova_sss_cons)
## Df Sum Sq Mean Sq F value Pr(>F)
## stress_lvl 3 1119 373.1 9.589 7.45e-06 ***
## Residuals 158 6147 38.9
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 4 observations deleted due to missingness
TukeyHSD(anova_sss_cons)
## Tukey multiple comparisons of means
## 95% family-wise confidence level
##
## Fit: aov(formula = sss_cons_sum ~ stress_lvl, data = s2_complete_data)
##
## $stress_lvl
## diff lwr upr p adj
## 2-1 5.0583004 1.643045 8.473556 0.0009910
## 3-1 6.2093301 2.622623 9.796038 0.0000784
## 4-1 6.3409091 2.642724 10.039094 0.0000937
## 3-2 1.1510297 -2.399360 4.701420 0.8344105
## 4-2 1.2826087 -2.380365 4.945582 0.7999659
## 4-3 0.1315789 -3.691754 3.954912 0.9997433
A one-way anova was performed with stress levels, four obseravtions were deleted due to missingness N=162. The effect was highky significant F(3.158)=9.5, p< .001 (p= 7.45e-06). This shows a large effect stress level which is explained by only 15.4% of variance on sss con sum. The Tukey post hoc test showed that participants in stress level have a significantly lower sss conservation score than all other levels. Level 2 (Mean diff= 6.21, p< .001) verses level 4 (Mean diff= 6.34, p< .001). Level 2, 3 and 4 did not differ from each other (all p> .79) showing a threshold effect where any increase above minimal stress (level 1) were associated with a higher conservation score, showing no further increase at a higher stress level.
#Sociability Sum
anova_sss_socio<-aov(sss_socio_sum ~ stress_lvl, data = s2_complete_data)
summary(anova_sss_socio)
## Df Sum Sq Mean Sq F value Pr(>F)
## stress_lvl 3 346.3 115.42 7.603 8.81e-05 ***
## Residuals 158 2398.5 15.18
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 4 observations deleted due to missingness
TukeyHSD(anova_sss_socio)
## Tukey multiple comparisons of means
## 95% family-wise confidence level
##
## Fit: aov(formula = sss_socio_sum ~ stress_lvl, data = s2_complete_data)
##
## $stress_lvl
## diff lwr upr p adj
## 2-1 2.0098814 -0.1234354 4.143198 0.0726964
## 3-1 3.0956938 0.8552805 5.336107 0.0024847
## 4-1 3.9331551 1.6231079 6.243202 0.0001064
## 3-2 1.0858124 -1.1319156 3.303540 0.5825329
## 4-2 1.9232737 -0.3647788 4.211326 0.1325556
## 4-3 0.8374613 -1.5507587 3.225681 0.7992646
A one-way anova showed a significant effect of stress levels on sss sociability sum F(3,158)=7.60, P< 8.81E-05. The Tukey post hoc test showed that participants in stress level 3 (Mean diff=3.10, p< .002, 95% CI [0.86, 534]) and level 4 (Mean diff=3.10, p=.002, 95% CI [1.62, 6.24])which show a significantly high socio sum than those in level 1. The difference between level 2 and level 1 was slim (Mean diff= 2.01, p= .73) no differences were found between levels 2, 3 and 4 (all p> .13).
Memory Performance
anova_labs<-aov(lab_stuff_score ~ stress_lvl, data = s2_complete_data)
summary (anova_labs)
## Df Sum Sq Mean Sq F value Pr(>F)
## stress_lvl 3 23.4 7.792 2.12 0.101
## Residuals 131 481.6 3.676
## 31 observations deleted due to missingness
TukeyHSD(anova_labs)
## Tukey multiple comparisons of means
## 95% family-wise confidence level
##
## Fit: aov(formula = lab_stuff_score ~ stress_lvl, data = s2_complete_data)
##
## $stress_lvl
## diff lwr upr p adj
## 2-1 1.01158301 -0.1648847 2.1880507 0.1185794
## 3-1 0.46872587 -0.7077419 1.6451936 0.7281723
## 4-1 0.01158301 -1.2381736 1.2613397 0.9999949
## 3-2 -0.54285714 -1.7355528 0.6498385 0.6377338
## 4-2 -1.00000000 -2.2650448 0.2650448 0.1728156
## 4-3 -0.45714286 -1.7221876 0.8079019 0.7831589
A one-way anova was performed to examine the effect of stress level on memory performance (lab stuff scores). The analysis included N=135 after 31 observations were deleted after missingness. The effects of stress levels were not significant, F(3,131)= 2.12, p= .101. The Tukey post hoc test showed no significant pairwise difference between stress levels (all p> .11). Descriptively, the largest difference was on level 2 and 4 (Mean diff= -1.00, p= .173 and 95% CI [-2.27, 0.27]). With level 2 showing a higher memory scores than level 4 but this did not show or reach any significance. Overall, stress levels did not significantly predict memory performance.
#Box plot for Memory vs Stress Scores
ggplot(s2_complete_data, aes(x = stress_lvl, y = lab_stuff_score))+
geom_boxplot(fill = "lightgreen")+
geom_jitter(alpha = 0.5)+
labs ( title = "Memory Performance by Stress Levels",
x = " Stress Levels",
y = " Labs Stuff Scores")+
theme_minimal()
## Warning: Removed 31 rows containing non-finite outside the scale range
## (`stat_boxplot()`).
## Warning: Removed 31 rows containing missing values or values outside the scale range
## (`geom_point()`).
The box plot shows that the distribution of memory scores across all four stress levels, N=135 and 31 missing observations. The black lines show the median, the green boxes show the interquartile range and the dots show individual scores. The median scores were 12.8 for level 1, 13.5 for level 2, 12.2 for level 3 and 11.9 for level 4. Level 1 showed a great variability (range ~ 8.5-17.5). Level 3 has two extrem outliers (~6.5) and level 2 showed a positive skew.
summary(lm(lab_stuff_score ~ poly(stress_lvl, 3),data = s2_complete_data))
##
## Call:
## lm(formula = lab_stuff_score ~ poly(stress_lvl, 3), data = s2_complete_data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.1714 -1.4371 0.2857 1.2857 5.2973
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 13.0930 0.1656 79.056 <2e-16 ***
## poly(stress_lvl, 3)1 -0.6033 2.1569 -0.280 0.780
## poly(stress_lvl, 3)2 -4.9185 2.1384 -2.300 0.023 *
## poly(stress_lvl, 3)3 2.3636 2.0988 1.126 0.262
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.917 on 131 degrees of freedom
## (31 observations deleted due to missingness)
## Multiple R-squared: 0.04629, Adjusted R-squared: 0.02445
## F-statistic: 2.12 on 3 and 131 DF, p-value: 0.1008
Memory performance was examined across all four stress levels. Descriptively, mean scores were highest at level 2 (Mean diff= 13.8) and low at levels 1 and 4 (Mean ~ 12.9). The one-way anova was not significant, F(3,131)= 2.12, p= .101 , R2=.05(with 31 observations deleted), as ploynomial trend analysis showed a significant negative quadratic trend, b=,-4,92, SE= 2.14, t(131)= -2,30, p=.23 with no significant linear(p=.780) and cubic (p=.262) components. This shows a inverted-U relationship with memory peaking at moderate stress, this is consistent with the Yerkes-Dadson’s law.
#Non-Linear: Trend Analysis Check
ggplot(s2_complete_data, aes(x = stress_lvl, y = lab_stuff_score))+
geom_point(alpha = 0.6, position = position_jitter(width = 0.15))+
geom_smooth(method = "lm", formula = y ~ poly(x,2, raw = TRUE), se = TRUE,
colour = "pink", fill = "green")+
stat_summary(fun = mean, geom = "point", colour = "blue",size = 4)+
scale_x_continuous(breaks = 1.4,labels = paste ("level", 1.4))+
labs( title = "Significant Quadratic Trend",
x = "Stress Levels",
y = "Lab Stuff Scores (Memory")+
theme_minimal()
## Warning: Removed 31 rows containing non-finite outside the scale range
## (`stat_smooth()`).
## Warning: Removed 31 rows containing non-finite outside the scale range
## (`stat_summary()`).
## Warning: Removed 31 rows containing missing values or values outside the scale range
## (`geom_point()`).
Although the overall one-way anova for labs stuff score was not
significant, F(3,131)= 2.12, P= .101, R2=.05. A planned polynimial
decomposition showed a significant trend, b=-4,92, SE= 2.14,
t(131)=-2.30 and p=.023. While linear, b=-0.60, p=.780 and cubic,
b=2.36, p=,262, the trends were not not significant. The negative
quadratic components show an inverted-U relationship between stress and
memory performance. This aligns with the Yerkes Dadson law, with a
perfornamce highest at a moderate stress (level 2)and lower at low
stress (level 1 and 4). The overall model remained non significant this
suggests that the effect is small and under powered (Adjusted R2=
.02).
Modeling Predictors
# Linear model with stress and other predictors
model_lab <- lm(lab_stuff_score ~stress_lvl+negative_affect_score, data = s2_complete_data)
summary(model_lab)
##
## Call:
## lm(formula = lab_stuff_score ~ stress_lvl + negative_affect_score,
## data = s2_complete_data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.0029 -1.1627 -0.0589 0.9871 4.9107
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 12.91196 0.68821 18.762 <2e-16 ***
## stress_lvl -0.02040 0.15353 -0.133 0.894
## negative_affect_score 0.01521 0.03632 0.419 0.676
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.954 on 132 degrees of freedom
## (31 observations deleted due to missingness)
## Multiple R-squared: 0.001502, Adjusted R-squared: -0.01363
## F-statistic: 0.09931 on 2 and 132 DF, p-value: 0.9055
When modeled linearly with negative affect as a covariant. Lab stuff scores were not significantly predicted F(2,132)=.10, p=.906, R2=.002. Neither stress level, b=-0.02 , SE=0.15, t(132)=0.13, p=.894 nor negative affect, b=0.02, SE=0.04, t(132)=0.42 and p=.679 were significant. The null linear effect hid a significant quardatic trend reported seperatly, t(133)=-2,30, p=.023 showing model misspecification when only linear terms are included.
#Short Recall Memory with
model_Short_Recall <- lm(short_recall_word ~ stress_lvl + negative_affect_score,
data = s2_complete_data)
summary(model_Short_Recall)
##
## Call:
## lm(formula = short_recall_word ~ stress_lvl + negative_affect_score,
## data = s2_complete_data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -10.7628 -1.7159 0.3889 1.8444 4.6973
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 17.44919 0.95653 18.242 <2e-16 ***
## stress_lvl -0.26839 0.20872 -1.286 0.200
## negative_affect_score -0.07664 0.05151 -1.488 0.139
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.905 on 159 degrees of freedom
## (4 observations deleted due to missingness)
## Multiple R-squared: 0.02265, Adjusted R-squared: 0.01035
## F-statistic: 1.842 on 2 and 159 DF, p-value: 0.1619
A multiple linear regression was performed to predict short call word scores from stress level and negative affect (N=162). The overall model was not significant F(2,159)=1.84, P=.162, R2=.02 and adjusted R2=.02. This shows that stress level and negative affect together explained only 2.3% variance. Therefore, neither stress level, b=-0.27, SE=0.21, t(159)=-1.29, p=.200 or negative affect, b=-0.08, SE=0.21, t(159)=-1.49, p=.139 were signifiant predictors.
#Delayed Recall Memory
model_delay <- lm(delay_recall_word ~ stress_lvl + negative_affect_score,
data = s2_complete_data)
summary(model_delay)
##
## Call:
## lm(formula = delay_recall_word ~ stress_lvl + negative_affect_score,
## data = s2_complete_data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -8.991 -2.973 0.527 2.651 7.226
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 12.01147 1.41150 8.510 3.21e-14 ***
## stress_lvl 0.71135 0.31372 2.267 0.025 *
## negative_affect_score -0.10824 0.07465 -1.450 0.149
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.017 on 133 degrees of freedom
## (30 observations deleted due to missingness)
## Multiple R-squared: 0.05393, Adjusted R-squared: 0.0397
## F-statistic: 3.791 on 2 and 133 DF, p-value: 0.02506
Delayed recall multiple regression predicting delayed word recall from stress level and negative affect was significant F(2,133)=3.79 p=.025, R2=.05 and adjusted R2=.04 with 30 observations were deleted due to missingness. Stress was a strong positive predictor b=0.71, SE=0.31, t(133)=2.27, p=.025 showed a high stressrelationship with better delayed recall. Negative affect was not a signifiant predictor, b=-0.11, SE= 0.07, t(1330=-1.45, p-.149.
Lineup Accuracy Data
clean_2$stress_num<-as.numeric(clean_2$stress_lvl)
model_lineup<-aov(stress_num ~ accuser, data = clean_2)
summary(model_lineup)
## Df Sum Sq Mean Sq F value Pr(>F)
## accuser 3 38.28 12.76 12.52 2.13e-07 ***
## Residuals 161 164.17 1.02
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
TukeyHSD(model_lineup)
## Tukey multiple comparisons of means
## 95% family-wise confidence level
##
## Fit: aov(formula = stress_num ~ accuser, data = clean_2)
##
## $accuser
## diff lwr upr p adj
## M-C 0.35448328 -0.1641008 0.8730673 0.2893656
## N-C -1.15785714 -1.7884107 -0.5273036 0.0000245
## R-C -0.05839768 -0.6137775 0.4969821 0.9928655
## N-M -1.51234043 -2.1612599 -0.8634210 0.0000001
## R-M -0.41288097 -0.9890280 0.1632661 0.2493452
## R-N 1.09945946 0.4207741 1.7781448 0.0002502
The one-way anova test performed was to examine whether stress level differed as a function of accuser lineup. Accuracy (clean_2, N= 165 after removing NA). There was a significant effect of accuracy on stress level F(3,161)=12.52, p<.001 and a variance of 19%. Tukey post hoc comparisons showed that N responses occured at a significantly lower stress levels than all other response types. Compared to C (Mean diff=-1.51, 95% CI [-1,79, -0.53], P<.001.Compared to M (Mean diff=-1.10, 95 % CI [0.42, 1.78], P<.001. No significant differences between C.M and R (all p>.24).
#For Proportion
table(clean_2$stress_lvl, clean_2$accuser)
##
## C M N R
## 1 12 6 16 9
## 2 16 11 9 10
## 3 15 13 0 10
## 4 13 17 0 8
chisq.test(table(clean_2$stress_lvl,clean_2$accuser))
##
## Pearson's Chi-squared test
##
## data: table(clean_2$stress_lvl, clean_2$accuser)
## X-squared = 34.556, df = 9, p-value = 7.133e-05
A chi-square test of independance showed a relationship between stress level and accuseror lineup accuracy (C,M,N,R) N=165. The relationship was significant X2(9)=34..56, P<.00. Cramers V=.26. The distribution showed that N reponses appeared exclusively at low stress level (level 1; N=165, 2%, level 2; N=9, 19,6%) and were absent at high levels of stress (level 3; N=0 , level 4; N=0). Concersly, the proportion of M responses increased from 14.05% at level 1 to 44.7% at level 4. This pattern was confirmed by the one-way anova test showing stress level differed significantly by accuracy , F(3,161)=12.52, p<.001. Tukey post hoc showed that N occured at a significanlty lower stress (Mean=1.4) than C, M and R (all Mean= 2.6-2.8, all p<.001) which did not differ from each other.
# Accuracy by stress level
ggplot(clean_2, aes(x = accuser, y = stress_num))+
geom_boxplot(fill = "purple") +
labs(title = "Line-up Accuracy by Stress Level",
x = "Accuser Accuracy",
y = "Stress Level")+
theme_minimal()
Stress level (1-4) differed significantly as a function of accuser
accuracy , F(3,161)=12.53, P<.001.The N responses occured exclusively
at low stress (median=1.2) and were absent at stress level 3 and 4.
Where M responses showed the highest median stress (median= 3.0) C and R
did nor differ significantly from each other.
Discussion
The present study examined the effects of experimentally induced stress on memory performance and eyewitness lineup decisions. Overall, the findings indicate that stress does not uniformly impair memory; its effects are task-dependent, following an inverted-U for encoding of peripheral detail, a linear facilitation for delayed recall, and a criterion shift for lineup decisions.
Lab Stuff Score When modeled linearly and controlling for negative affect, stress did not predict memory for lab details, F=0.10, p=.906, R2=.002. However, this null linear effect masked a significant quadratic trend. A polynomial model revealed a significant negative quadratic term, b =-4.92, t(131)=-2.30, p=.023, with medium effect size (R2=.046). Performance peaked at Stress Level 2 (Mean=13.8) and declined at both low (Level 1: Mean=12.9) and high stress (Level 4: Mean=12.9).This pattern is consistent with the Yerkes-Dodson law and attentional narrowing theory (Easterbrook, 1959; Deffenbacher, 1983). At low stress, arousal is insufficient for optimal input. At high stress, attention narrows to central, threat-relevant cues at the expense of peripheral details such as the lab environment. This explains why a linear model averaging across this curve yielded b=-0.02, p=.894 - the positive and negative slopes cancel out. The finding that negative affect did not predict lab scores, b=0.02, p=.676, suggests the effect is driven by physiological arousal rather than mood valence.
Short vs. Delayed Recall: Dissociation This dissociation is theoretically important and aligns with memory consolidation literature. Acute stress and cortisol release during encoding enhance long-term consolidation via the amygdala-hippocampal interactions (McGaugh, 2000; Schwabe et al., 2012), while leaving immediate retrieval relatively unaffected or impaired. The lab task, which tested incidental peripheral memory, is vulnerable to attentional narrowing, whereas the recall task tested central, rehearsed information which benefits from stress-induced consolidation. Thus stress impairs what you attend to at but enhances how strongly attended information is stored.
Line-up Accuracy This is not a memory accuracy effect per se, but a decision-criterion shift. Under low stress, participants maintained a conservative criterion and were able to withhold an identification. Under high stress, this ability collapsed, consistent with increased liberal responding and reduced monitoring under high arousal (Weber & Perfect, 20120. Critically. C remained stable (∼28-39% across levels), meaning high stress did not improve true hits - it only converted conservative N responses into false M responses. For eyewitness testimony, this is highly forensically relevant: high stress does not make witnesses more accurate, it makes them less willing to say “I don’t know,” inflating false identifications.
Limitations and Future Directions
First, lab_stuff_score had 31 missing cases, reducing power (N=135 vs N=165 for lineup). Second, stress was manipulated as a 4-level ordinal variable; future research should include physiological manipulation checks (for example, heart rate, cortisol) to verify linearity of the manipulation.
Practical Study Implications
In terms of the legal system, the present data suggest that high stress is problematic: it reduces memory for peripheral context that can corroborate testimony. Meaning it eliminates the witness’s ability to correctly withhold an identification while increasing the risk of foil picks by over 30 percentage points from low to high stress. Training interviewers to emphasize that “not present” is an acceptable answer may be especially critical under high-stress conditions.
Conclusion
The assignment shows that the relationship between stress and eyewitness memory cannot be reduced to a simple “stress impairs memory” statement.Rather, the effect of stress is qualitatively different depending on what is being tested and when.Two distinct patterns emerged, firstly, memory for peripheral lab details followed a significant inverted-U function, p=.023, with optimal performance at moderate stress and impairment at both low and high stress. When forced into a linear model, this effect disappeared entirely, F=0.10, p=.906, highlighting the importance of modelling non-linear stress effects. Secondly, a dissociation was observed between immediate and delayed recall, short-term recall was unaffected by stress, p=.162. Whereas delayed recall showed a significant linear improvement, b=0.71, p=.025, supporting a consolidation-enhancement mechanism.
In the context of forensic practice, the implication is clear: under high stress, witnesses do not become less accurate in the sense of forgetting the perpetrator, but they become less able to say “I don’t know,” which is precisely the condition that produces wrongful convictions.