pd <- position_dodge(0.3) 

## Gives count, mean, standard deviation, standard error of the mean, and confidence interval (default 95%).
##   data: a data frame.
##   measurevar: the name of a column that contains the variable to be summariezed
##   groupvars: a vector containing names of columns that contain grouping variables
##   na.rm: a boolean that indicates whether to ignore NA's
##   conf.interval: the percent range of the confidence interval (default is 95%)
summarySE <- function(data=NULL, measurevar, groupvars=NULL, na.rm=FALSE,
                      conf.interval=.95, .drop=TRUE) {

    # New version of length which can handle NA's: if na.rm==T, don't count them
    length2 <- function (x, na.rm=FALSE) {
        if (na.rm) sum(!is.na(x))
        else       length(x)
    }

    # This does the summary. For each group's data frame, return a vector with
    # N, mean, and sd
    datac <- ddply(data, groupvars, .drop=.drop,
      .fun = function(xx, col) {
        c(N    = length2(xx[[col]], na.rm=na.rm),
          mean = mean   (xx[[col]], na.rm=na.rm),
          sd   = sd     (xx[[col]], na.rm=na.rm)
        )
      },
      measurevar
    )

    # Rename the "mean" column    
    datac <- rename(datac, c("mean" = measurevar))

    datac$se <- datac$sd / sqrt(datac$N)  # Calculate standard error of the mean

    # Confidence interval multiplier for standard error
    # Calculate t-statistic for confidence interval: 
    # e.g., if conf.interval is .95, use .975 (above/below), and use df=N-1
    ciMult <- qt(conf.interval/2 + .5, datac$N-1)
    datac$ci <- datac$se * ciMult

    return(datac)
}

# Multiple plot function
#
# ggplot objects can be passed in ..., or to plotlist (as a list of ggplot objects)
# - cols:   Number of columns in layout
# - layout: A matrix specifying the layout. If present, 'cols' is ignored.
#
# If the layout is something like matrix(c(1,2,3,3), nrow=2, byrow=TRUE),
# then plot 1 will go in the upper left, 2 will go in the upper right, and
# 3 will go all the way across the bottom.
#
multiplot <- function(..., plotlist=NULL, file, cols=1, layout=NULL) {
  library(grid)

  # Make a list from the ... arguments and plotlist
  plots <- c(list(...), plotlist)

  numPlots = length(plots)

  # If layout is NULL, then use 'cols' to determine layout
  if (is.null(layout)) {
    # Make the panel
    # ncol: Number of columns of plots
    # nrow: Number of rows needed, calculated from # of cols
    layout <- matrix(seq(1, cols * ceiling(numPlots/cols)),
                    ncol = cols, nrow = ceiling(numPlots/cols))
  }

 if (numPlots==1) {
    print(plots[[1]])

  } else {
    # Set up the page
    grid.newpage()
    pushViewport(viewport(layout = grid.layout(nrow(layout), ncol(layout))))

    # Make each plot, in the correct location
    for (i in 1:numPlots) {
      # Get the i,j matrix positions of the regions that contain this subplot
      matchidx <- as.data.frame(which(layout == i, arr.ind = TRUE))

      print(plots[[i]], vp = viewport(layout.pos.row = matchidx$row,
                                      layout.pos.col = matchidx$col))
    }
  }
}
data = read.csv("../covid19_study1_clean_wide.csv", stringsAsFactors = FALSE)

## reverse coding
cols_data = c( "beliefs_norms_1")
data[,cols_data] = lapply(cols_data,  function(x) 8 - data[, x])

 # windsorize items
for (i in c(1:ncol(data))){
   if (is.numeric(data[,i])){
     m = mean(data[,i])
     s = sd(data[,i])
     if (length(which(data[,i] > m + 3*s)) > 0){
       data[which(data[,i] > m + 3*s),i] = m + 3*s
     }
    if (length(which(data[,i] < m - 3*s)) > 0){
       data[which(data[,i] < m - 3*s),i] = m + 3*s
     }
   }
 }
 

# get norm and intention ratings

data$norm_town_socialdis = scale(rowMeans(data[,c("norms_town1_2",  "norms_town1_4", "norms_town1_6", "norms_town1_10")]))
data$norm_close_socialdis = scale(rowMeans(data[,c("norms_close1_2","norms_close1_4", "norms_close1_6","norms_close1_10")]))
data$intention_socialdis = scale(rowMeans(data[,c("intentions1_2","intentions1_4","intentions1_6","intentions1_10")]))
data$self_beliefs = scale(rowMeans(data[,c("beliefs_safe_self_1","beliefs_safe_self_2","beliefs_safe_self_3","beliefs_safe_self_4","beliefs_safe_self_5")]))
data$other_beliefs = scale(rowMeans(data[,c("beliefs_safe_others_1","beliefs_safe_others_2","beliefs_safe_others_3",
                                      "beliefs_safe_others_4","beliefs_safe_others_5",
                                      "beliefs_safe_others_6","beliefs_safe_others_7","beliefs_safe_others_8")]))
data$norm_beliefs = scale(rowMeans(data[,c("beliefs_norms_1","beliefs_norms_2","beliefs_norms_3","beliefs_norms_4")]))

data$independence = scale(rowMeans(data[,c("selfconstrual_1","selfconstrual_2","selfconstrual_3","selfconstrual_4")]))
data$interdependence = scale(rowMeans(data[,c("selfconstrual_5","selfconstrual_6","selfconstrual_7","selfconstrual_8")]))

df_all = data
  
df_norm = data[which(data$condition %in% c( "message control", "norm")), ]

For any questions regarding this report, please contact Rui Pei at rui.pei@asc.upenn.edu.

Executive summary

In this study, we examined whether messages that incorporate descriptive norms would be effective at influencing people’s covid-related cognitions. Overall, the data did not provide evidence that norm-based messages would be perceived more favorabily and lead to increased social norm perception. We speculate that our norm manipulation may not have been salient enough. As seen in the example below, messages across conditions shared the same image and part of the description. As participants may have paid more attention on the images, it could be that our manipulation was not strong enough,

On the other hand, the analyses supported Hypothesis 3, showing a positive association between interdependence and perceived social distancing norms.

Introduction

The opinions and behaviors of peers are potent factors influencing human decision making (Cialdini & Goldstein, 2004). As such, a number of studies have demonstrated that descriptive norm-based messages can be effective at influencing individuals’ behavior through changing their perceived norms. Such health campaigns have been carried out in the U.S., Canada, and the UK, and targeted a number of risky and health behaviors, such as drug use, alcohol consumption, and tobacco use (e.g. Arbour-Nicitopoulos, Kwan, Lowe, Taman, & Faulkner, 2010; Cislaghi & Heise, 2018; Donaldson et al., 1994; Larimer & Neighbors, 2003; Liu & Shi, 2019; Neighbors et al., 2004). In addition to descriptive norms, other persuasive appeals (such as autonomy and humor) have shown to be effective at influencing people’s health-related beliefs and behaviors.

The aim of this study was to examine the effectiveness of norm-based covid-19 social distancing messages (e.g. messages that highlight people staying home and avoiding crowds) in increasing people’s perceived social distancing norms. In addition, we investigated the role of interdependent self-construal in moderating the effects of our experimental manipulations (norms/autonomous/humor) on perceived norms.

Methods

Full methods for this study can be found at https://cnlab.github.io/covid19-message-framing/.

Preregistration for this study can be found at https://osf.io/xg4sa.

Other data reports related to this project can be found at https://cnlab.github.io/covid19-message-framing/.

865 participants were recruited from Amazon Mechanical Turk platform (mean age = 37.65, SD age = 11). Specific breakdown of participant demographics by condition can be found in Table 1.

Table 1
condition N age sd % female
autonomous 166 37.57 11.02 43.37
message control 179 39.53 12.04 45.25
mocking 171 36.26 11.19 46.20
no message control 179 37.61 11.31 49.72
norm 170 37.21 10.92 38.24

Primary Results

Hypothesis 1.

Descriptive norm messages will be rated as more favorable and socially relevant compared to control messages.

In our preregistration, we stated that we would test if age was significantly associated with each respective DV, and if so, we would include age as a control variable in our models. Age was significantly associated with favoribility but not social relevance (favorability: b = 0.01, 95% CI = [0.01, 0.02], p < 0.001 ; social relevance: b = 0.01, 95%CI = [-0.0003, 0.01], p = 0.06). Therefore, we included age as a control variable for the model predicting message favorability, but not social relevance.

To test our hypotheses that descriptive norm messages will be perceived as more favorable and socially relevant, we constructed two mixed effect multilevel regression models in which condition was treated as a fixed effect, and participant and message were treated as random effects. For the model predicting message favorability, we included age as a fixed effect control as well. Our results showed no evidence that messages that highlight descriptive norms were rated as more favorable (b = 0.04, 95%CI = [-0.14, 0.22], p = 0.67) nor socially relevant (b = 0.06, 95%CI = [-0.09, 0.22], p = 0.43) compared to control messages. As such, hypothesis 1 was not supported.

favorability social relevance
norm (vs. control) 0.04     0.06 
(0.09)    (0.08)
age 0.01 ***     
(0.00)        
N 1739        1739    
N (SID) 349        349    
N (question) 15        15    
AIC 3551.44     3875.18 
BIC 3584.21     3902.49 
R2 (fixed) 0.03     0.00 
R2 (total) 0.73     0.58 
*** p < 0.001; ** p < 0.01; * p < 0.05.

Hypothesis 2.

Compared to participants in the control condition, participants in the descriptive norm condition will show higher levels of perceived norms regarding social distancing behavior in people in their town/city.

We constructed a multiple linear regression model to test our hypothesis that participants in descriptive norm condition will show higher levels of perceived social distancing norms. Since age was significantly associated with participants perceived social distancing norms in their town/city (b = 0.01, 95%CI = [0.004, 0.02], p = 0.004), age was included as a covariate in our model. Our results demonstrated that there were no significant differences in perceived norm between participants in norm versus message control condition (b = -0.04, 95%CI = [-0.26, 0.17], p = 0.68). As such, hypothesis 2 was not supported.

F(2,346) 4.10
R² 0.02
Adj. R² 0.02
Est. 2.5% 97.5% t val. p
(Intercept) -0.46 -0.86 -0.06 -2.26 0.02
conditionnorm -0.04 -0.26 0.17 -0.41 0.68
age 0.01 0.00 0.02 2.78 0.01
Standard errors: OLS

Hypothesis 3.

In general, there will be a positive association between participants’ interdependence level and their perceived social distance norms of people in their town / city (for participants in the no message control condition).

We constructed a multiple linear regression model to test our hypothesis that for participants in no message control condition, there will be a positive association between interdependent self-construal and their perceived social distance norms in their town/city. Since age was not significantly associated with perceived social distancing norms for participants in no message control condition (b = 0.01, 95%CI = [-0.003, .02], p = 0.13), age was not included as a covariate in this model. Our results demonstrated a significant association between participants’ interdependence self-construal and perceived norms in their town/city (b = 0.20, 95%CI = [0.07, 0.33], p = .002). As such, hypothesis 3 was supported.

F(1,177) 9.62
R² 0.05
Adj. R² 0.05
Est. S.E. t val. p
(Intercept) 0.06 0.07 0.85 0.40
interdependence 0.20 0.07 3.10 0.00
Standard errors: OLS

Hypothesis 4.

Participants’ self-report ratings of interdependent self-construal will moderate the relationship between experimental condition (norm/humor/autonomy vs. control) and their perceived social distancing norms of people in their town/city.

To test hypothesis 4, a multiple linear regression model was constructed to investigate hypothesis 4. Since age was a significant predictor for perceived social norms (b = 0.01, 95%CI = [0.001, 0.02], p = 0.02) in this sample (all participants excluding the participants in the no message control condition), age was included as a covariate in this model. Our results suggested that there was no significant interaction effect between interdependence and experimental condition in predicting perceived social norms (b = -0.10, 95%CI = [-0.29, 0.08], p = 0.27).

We also constructed a separate model to examine Hypothesis 3 (association between interdependence and norm perception) in this bigger sample. In this model, we observed a significant main effect of interdependence on perceived norm (b = 0.14, 95%CI = [0.07, 3.66], p < .001), replicating Hypothesis 3 in this bigger sample that includes participants in experimental as well as message control conditions. In sum, participant’s interdependence self-construal was positively associated with their perceived social distancing norms. Participants with higher interdependent self-construal on average reported more people in their town/city practising social distancing. This relationship was not moderated by our messaging manipulation.

interaction model main effect
condition x interdependence -0.10           
(0.09)          
experimental (vs. control) -0.08    -0.08    
(0.09)   (0.09)   
interdependence 0.22 ** 0.14 ***
(0.08)   (0.04)   
age 0.09 *  0.09 *  
(0.04)   (0.04)   
N 685       685       
R2 0.03    0.03    
All continuous predictors are mean-centered and scaled by 1 standard deviation. *** p < 0.001; ** p < 0.01; * p < 0.05.

Secondary analyses

Analyses 1-4.

The effects of descriptive norm manipulation on message ratings: self relevance, self motivation, other motivation, and sharing intention.

Analyses 1 - 4 are related to the message level ratings: self relevance (“this message is relevant to me”), self motivation (“this message motivates me to avoid physical contact with others”), other motivation (“this message would motivate people I know to avoid physical contact with others”), and sharing intentio (“I would share this message on social media”). Based on pilot data, descriptive norm condition (relative to message control condition) had a smaller effect on these message ratings (self relevance, self motivation, other motivation, and sharing intention). We therefore proposed these effects as secondary analyses.

The results showed that compared to control messages, messages in the descriptive norm condition did not receive significantly higher ratings in self relevance, self motivation, other motivation, and sharing intention.

self relevance self motivation other motivation sharing intention
(Intercept) -0.28   -0.35 *  -0.34 *  -0.28 
(0.15)  (0.15)   (0.14)   (0.17)
conditionnorm 0.06   0.02    0.04    0.01 
(0.08)  (0.08)   (0.07)   (0.09)
age 0.01 * 0.01 ** 0.01 ** 0.01 
(0.00)  (0.00)   (0.00)   (0.00)
N 1739      1739       1739       1739    
N (SID) 349      349       349       349    
N (question) 15      15       15       15    
AIC 3998.94   4189.32    4267.87    3797.92 
BIC 4031.70   4222.09    4300.64    3830.69 
R2 (fixed) 0.01   0.01    0.01    0.01 
R2 (total) 0.57   0.51    0.45    0.69 
*** p < 0.001; ** p < 0.01; * p < 0.05.

Analyses 5.

Investigate the effect of descriptive norm condition on participant’s social distancing intentions.

We investigated whether messages that incorporate descriptive norms would be effective at shifting people’s social distancing intentions compared to control messages. Our data showed that compared to control messages, participants in descriptive norm condition did not show higher intentions to social distance (b = -0.10, 95%CI = [-0.32, 0.11], p = 0.35).

F(2,346) 1.90
R² 0.01
Adj. R² 0.01
Est. 2.5% 97.5% t val. p
(Intercept) -0.25 -0.65 0.15 -1.25 0.21
conditionnorm -0.10 -0.32 0.11 -0.94 0.35
age 0.01 -0.00 0.02 1.60 0.11
Standard errors: OLS

Analysis 6.

The effect of descriptive norm condition on participant’s COVID-19 beliefs.

We investigated whether messages that incorporate descriptive norms would be effective at shifting people’s beliefs about social distancing. Three types of beliefs were investigated: the belief that practising social distancing will keep myself safe (“self beliefs”), beliefs that practising social distancing will keep others safe (“other beliefs”), and beliefs that other people will approve my decision of social distancing (“norm belief”). Our data showed that compared to control messages, participants in descriptive norm condition did not show higher levels of self beliefs (b = -0.06, 95%CI = [-0.26, 0.14], p = 0.55), other beliefs (b = -0.08, 95%CI = [-0.28, 0.11], p = 0.40), or norm beliefs (b = 0.15, 95%CI = [-0.06, 0.36], p = 0.17).

self beliefs other beliefs norm beliefs
(Intercept) -0.05  -0.07  -0.34 
(0.19) (0.19) (0.20)
conditionnorm -0.06  -0.08  0.15 
(0.10) (0.10) (0.11)
age 0.00  0.00  0.01 
(0.00) (0.00) (0.00)
N 349     349     349    
R2 0.00  0.00  0.01 
*** p < 0.001; ** p < 0.01; * p < 0.05.

Analysis 7.

The effect of descriptive norm condition on participant’s social distancing norms of their friends and family.

We investigated whether messages that incorporate descriptive norms would be effective at shifting people’s perceived social distancing norms of their friends and family. Since age was a significant predictor of perceived norms in friends and family (b = 0.02, 95%CI = [0.01, 0.03], p < .001), we included age as a covariate in our multiple regression model. Our data showed that compared to control messages, participants in descriptive norm condition did not show higher levels of perceived norms in their close friends and family (b = 0.07, 95%CI = [-0.14, 0.28], p = 0.53).

F(2,345) 7.97
R² 0.04
Adj. R² 0.04
Est. 2.5% 97.5% t val. p
(Intercept) -0.73 -1.12 -0.34 -3.69 0.00
conditionnorm 0.07 -0.14 0.28 0.63 0.53
age 0.02 0.01 0.03 3.99 0.00
Standard errors: OLS

Analysis 8-10.

Whether age moderates the effect of condition on message social relevance, favorability, and perceived norms in town/city.

We tested whether age moderates the effect of condition on message social relevance and favorability using multilevel mixed effect models in which participant and message were included as random effects. The results demonstrated no significant interaction between age and condition predicting message favorability relevance (b = -0.003, 95%CI = [-0.19, 0.01], p = 0.72) nor message social relevance (b = -0.003, 95%CI = [-0.02, 0.01], p = 0.65). In addition, we examined whether the effects of experimental condition on participants’ perceived norms in town/city was moderated by age. The results showed no significant interaction between age and condition predicting perceived social norms (b = 0.001, 95%CI = [-0.02, 0.02], p = 0.91).

favorability social relevance social norm in town/city
(Intercept) -0.537 *  -0.297  -0.442  
(0.220)   (0.190) (0.263) 
conditionnorm 0.149    0.199  -0.084  
(0.319)   (0.275) (0.384) 
age 0.015 ** 0.008  0.013 *
(0.005)   (0.005) (0.006) 
conditionnorm:age -0.003    -0.003  0.001  
(0.008)   (0.007) (0.010) 
N 1739        1739      349      
N (SID) 349        349            
N (question) 15        15            
AIC 3561.135    3892.799  1011.819  
BIC 3599.362    3931.027  1031.094  
R2 (fixed) 0.027    0.009        
R2 (total) 0.731    0.577  0.023  
*** p < 0.001; ** p < 0.01; * p < 0.05.

Analysis 11.

The effects of pronouns (“I”, “we”, “most people”) in descriptive messages on message ratings.

All our descriptive norm messages were written using one of the following three pronouns: “I”, “we”, “most people”. For instance:

  • “I” message: Social distancing is critical for reducing exposure to #covid19. I am staying home to stop hundreds of others from being infected.

  • “We” message: Social distancing is critical for reducing exposure to #covid19. We are staying home to stop hundreds of others from being infected.

  • “Most people” message: Social distancing is critical for reducing exposure to #covid19. Most people are staying home to stop hundreds of others from being infected.

We investigated if different wording of descriptive norm messages (using either “I”, “we”, or “most people”) has an effect on participants’ ratings on message favorability, motivation, and relevance, compared to the control condition. This was tested in participants in the descriptive norm and message control conditions. No significant effects were observed in our data.

favorability other motivation self motivation self relevance social relevance share
(Intercept) -0.50 **  -0.38 *  -0.40 ** -0.26   -0.24  -0.34 
(0.16)    (0.15)   (0.15)   (0.15)  (0.15) (0.17)
message_typemost 0.08     0.12    0.10    0.05   0.08  0.12 
(0.07)    (0.12)   (0.10)   (0.09)  (0.07) (0.09)
message_typewe 0.04     0.08    0.07    -0.01   0.05  0.07 
(0.07)    (0.12)   (0.10)   (0.09)  (0.07) (0.09)
age 0.01 *** 0.01 ** 0.01 ** 0.01 * 0.01  0.01 
(0.00)    (0.00)   (0.00)   (0.00)  (0.00) (0.00)
N 1739        1739       1739       1739      1739     1739    
N (SID) 349        349       349       349      349     349    
N (question) 15        15       15       15      15     15    
AIC 3556.80     4270.81    4192.74    4003.96   3888.40  3801.36 
BIC 3595.03     4309.04    4230.97    4042.19   3926.62  3839.59 
R2 (fixed) 0.03     0.01    0.02    0.01   0.01  0.01 
R2 (total) 0.73     0.46    0.51    0.57   0.58  0.69 
*** p < 0.001; ** p < 0.01; * p < 0.05.