Code
library(tidyverse)
library(readxl)
# read in your data here using read.csv
mydf <- read_excel("1901_2026_s1_data.xlsx")The Influence of Baseline Side-effect Expectancy and Positive Social Modelling on Severity of Nocebo side Effects
Executive Summary
Baseline side-effect expectancy showed a small positive association with primary nocebo symptom change. Positive social modelling reduced overall symptom severity and variability, although it did not clearly weaken the expectancy–symptom relationship. This suggests that expectancy contributes to nocebo responding, but intervention effects are minimal and context-dependent.
Evidence
Exploratory Data Analysis
The data were sourced from Cosette Saunders’ research on nocebo effects and positive social modelling. The dataset contains 161 observations (participants) and 80 variables. This analysis focused on two key numeric variables: expect_sideeffect, measuring baseline expectancy of adverse effects from study participation, and primary, a derived variable representing the primary side-effect outcome, calculated as the active-minus-baseline difference score for headache and dizziness.
The dataset is structured in wide format, with baseline and post-treatment measures recorded in the same row for each participant. This structure is appropriate for examining change over time and modelling the relationship between expectancy and symptom severity. While the variables used here are numeric, some other dataset variables, such as treatment condition, required reclassification as factors in R.
Limitations
Several limitations should be acknowledged. The sample is predominantly young, with a median age of 19 years, limiting generalisability beyond university-aged participants. Some physiological variables are also stored as character strings rather than numeric values, and some derived variables require careful interpretation to ensure consistency with the original survey items.
Assumptions
The analysis assumes independent observations, consistent variable coding, and valid construction of primary. Data cleaning involved inspecting variable types, checking for missing or inconsistent values, and converting variables into appropriate formats.
Data Analysis
Baseline side-effect expectancy was a statistically significant predictor of primary symptom change, but the magnitude of this effect was small. In figure 1, each 1-point increase in expected side effects was associated with an average 0.025-unit increase in primary symptom change. Although statistically significant, this effect is minimal. From figure 1, the fitted line slopes only slightly upward, while most observations remain between 0 and 2 units of symptom change across the expectancy range, with several outliers such as at 6 and -5, suggesting that expectancy contributes to nocebo responding, but accounts for only a limited proportion of symptom variation, consistent with Rooney et al. (2023).
The coefficient for intervention_binNo Int was 1.201, indicating that at an expectancy value of zero, the no-intervention group had, on average, 1.20 units higher primary symptom change. This implies that positive social modelling was associated with a lower baseline level of nocebo symptom severity. However, since the intervention-group slope was 0.025, the corresponding no-intervention slope was approximately −0.007. This pattern is visible in Figure 2: the intervention line trends slightly upward, whereas the no-intervention line is nearly flat to slightly negative such that the intervention appears to reduce overall symptom levels, but it does not weaken the expectancy–symptom relationship in the way originally hypothesised. Broader literature shows that modelling influences symptom reporting even when effects vary in size and direction across contexts (Faasse et al., 2015).
Further, figure 3 shows that though positive social modelling can reduce the extremities of nocebo responses—as IQR(int)=1.25 whilst IQR(No int)=2.00—it does not eliminate the influence of expectancy. However, this model explains only a small proportion of overall variance . However, the model explains only a small proportion of the overall variance (R² = .049, adjusted R² = .031), so the findings should be interpreted cautiously. This interpretation is consistent with evidence showing that nocebo effects arise from multiple interacting risk factors, including expectations, learning processes, and contextual influences (Grosso et al., 2024) rather than expectations alone.
Code
# you can enter your code to make your graphs in code chunks like this one
# the output will be displayed in your html
library(tidyverse)
mydf <- mydf %>%
mutate(
intervention_bin = factor(intervention_bin),
gender = factor(gender)
)
#this code chunk fig 1 shows the overall relationship between expectancy and symptom change.
ggplot(mydf, aes(x = expect_sideeffect, y = primary)) +
geom_jitter(width = 1.5, height = 0.08, alpha = 0.55, size = 2) +
geom_smooth(method = "lm", se = TRUE, linewidth = 1) +
labs(
title = "Figure 1. Baseline Side-Effect Expectancy \nand Primary Nocebo Symptom Severity",
x = "Baseline expectation of side effects",
y = "Primary symptom change"
) +
theme_minimal(base_size = 13) +
theme(
plot.title = element_text(face = "bold"),
panel.grid.minor = element_blank()
)`geom_smooth()` using formula = 'y ~ x'
Code
#this plot shows the relationship by intervention condition
ggplot(mydf, aes(x = expect_sideeffect, y = primary, colour = intervention_bin)) +
geom_jitter(width = 1.5, height = 0.08, alpha = 0.5, size = 2) +
geom_smooth(method = "lm", se = FALSE, linewidth = 1.1) +
labs(
title = "Figure 2. Expectancy and Primary Symptom Change \nby Intervention Condition",
x = "Baseline expectation of side effects",
y = "Primary symptom change",
colour = "Intervention"
) +
theme_minimal(base_size = 13) +
theme(
plot.title = element_text(face = "bold"),
panel.grid.minor = element_blank(),
legend.position = "top"
)`geom_smooth()` using formula = 'y ~ x'
Code
#this plot shows the distribution of symptom change by intervention group
ggplot(mydf, aes(x = intervention_bin, y = primary, fill = intervention_bin)) +
geom_boxplot(alpha = 0.6, outlier.shape = NA, width = 0.55) +
geom_jitter(width = 0.12, alpha = 0.45, size = 1.8) +
labs(
title = "Figure 3. Distribution of Primary Nocebo Symptom Severity \nby Intervention Group",
x = "Intervention condition",
y = "Primary symptom change",
fill = "Intervention"
) +
theme_minimal(base_size = 13) +
theme(
plot.title = element_text(face = "bold"),
panel.grid.minor = element_blank(),
legend.position = "none"
)
*** Linear Regression Model Statistical Summary (fig 1):
Code
#the below code allows us to see the linear regression mode
model1 <- lm(primary ~ expect_sideeffect * intervention_bin, data = mydf)
summary(model1)
Call:
lm(formula = primary ~ expect_sideeffect * intervention_bin,
data = mydf)
Residuals:
Min 1Q Median 3Q Max
-5.2923 -0.9552 -0.2931 0.8720 5.2248
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.04209 0.34347 0.123 0.90262
expect_sideeffect 0.02502 0.01219 2.053 0.04176
intervention_binNo Int 1.20113 0.42279 2.841 0.00509
expect_sideeffect:intervention_binNo Int -0.03222 0.01376 -2.341 0.02048
(Intercept)
expect_sideeffect *
intervention_binNo Int **
expect_sideeffect:intervention_binNo Int *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 1.452 on 157 degrees of freedom
Multiple R-squared: 0.04919, Adjusted R-squared: 0.03103
F-statistic: 2.708 on 3 and 157 DF, p-value: 0.04716*** Box Plot Five Number Summary:
Code
#i want the five number summary for the box plots:
mydf %>%
group_by(intervention_bin) %>%
summarise(
min = min(primary, na.rm = TRUE),
Q1 = quantile(primary, 0.25, na.rm = TRUE),
median = median(primary, na.rm = TRUE),
Q3 = quantile(primary, 0.75, na.rm = TRUE),
max = max(primary, na.rm = TRUE)
)# A tibble: 2 × 6
intervention_bin min Q1 median Q3 max
<fct> <dbl> <dbl> <dbl> <dbl> <dbl>
1 Int -5 0 0 1.25 4
2 No Int -2 0 0 2 6Acknowledgements
Group Meetings:
- 06/04/2026 from 10am-4pm. All members in attendance. Zichen Wang worked on EDA, Julian Thomas worked on researching context and scientiic literature surrounding the research question, Omar and Aaryash Bharadwaj worked on producing figures for the data analysis. Online search engines and ed-lessons were used for this. -08/04/2026 from 4pm-8pm. All members in attendance. Zichen and Aaryash worked on writing the analysis of the data and Julian and Omar worked on refining the details with the graphs and the bibliography. Same resources were used as previous meeting.
AI Usage Statement:
Chat GPT (GPT 4.5 thinking model) was used. The publisher is OpenAI.(https://chatgpt.com/auth/login/?next=%2Fg%2Fg-p-69b64308ef8c819182b528e032bb1eab-data1901%2Fproject) The tool was used in order to help diagnose issues with code, proofread the final quarto document and also to find relevant scientific literature.
Adherence to Ethical and Professional Standard:
One of the shared professional values of truthfulness and integrity has been adhered to as we made our techniques for manipulating data, creating graphs and reclassifying data reproducible and honest. We maintained the ethical principles of ‘pursuing objectivity’ by ensuring that all data analysis was unbiased and transparent rather than for an ad-hoc purpose.
References:
Faasse, K., Grey, A., Jordan, R., Garland, S., & Petrie, K. J. (2015). Seeing is believing: Impact of social modeling on placebo and nocebo responding. Health Psychology, 34(8), 880–885. https://doi.org/10.1037/hea0000199
Grosso, F., Barbiani, D., Cavalera, C., Volpato, E., & Pagnini, F. (2024). Risk factors associated with nocebo effects: A review of reviews. Brain, Behavior, & Immunity - Health, 38, 100800. https://doi.org/10.1016/j.bbih.2024.100800
Rooney, T., Sharpe, L., Todd, J., Richmond, B., & Colagiuri, B. (2023). The relationship between expectancy, anxiety, and the nocebo effect: A systematic review and meta-analysis with recommendations for future research. Health Psychology Review, 17(4), 550–577. https://doi.org/10.1080/17437199.2022.2125894
Saunders, C., Tan, W., Faasse, K., Colagiuri, B., Sharpe, L., & Barnes, K. (2024). The effect of social learning on the nocebo effect: A systematic review and meta-analysis with recommendations for the future. Health Psychology Review, 18(4), 934–953. https://doi.org/10.1080/17437199.2024.2394682
Saunders, C., Tan, W., Ng, D., Burchett, A., McNair, N., & Colagiuri, B. (2025). Positive social modeling attenuates nocebo side effects. Annals of Behavioral Medicine, 59(1), kaaf048. https://doi.org/10.1093/abm/kaaf048
Quinn, V., Pearson, S., Huynh, A., Nicholls, K., Barnes, K., & Faasse, K. (2023). The influence of video-based social modelling on the nocebo effect. Journal of Psychosomatic Research, 165, 111136.