# these two commands clear the R workspace rm(list=ls()) cat("\014") # specify the directory that the csv file is in and the csv file name # setwd("Statistics HRI Guide/") jibo_is_data <- read.csv("jibo_survival_experiment_data_independent_samples.csv", header=T) # indicate missing data jibo_is_data[jibo_is_data == "NA"] <- NA # for any variables that are categorical, specify them as factors jibo_is_data$condition_2_Lin_3_Lout <- factor( jibo_is_data$condition_2_Lin_3_Lout, levels = 2:3, labels = c("liaison_ingroup", "liaison_outgroup") ) jibo_is_data$condition_1_Rbc_0_Rnobc <- factor( jibo_is_data$condition_1_Rbc_0_Rnobc, levels = 0:1, labels = c("backchanneling_robot", "silent_robot") ) jibo_is_data$produced_more_than_100_bcs <- factor( jibo_is_data$produced_more_than_100_bcs, levels = 0:1, labels = c("less_than_100_bcs_produced", "more_than_100_bcs_produced") ) # ------------------------------------------------------------------------------------------------# # one categorical IV and one categorical DV --> Chi Square Test of Independence # # - we will test whether the between subjects condition of whether the robot is backchanneling # # or not (condition_1_Rbc_0_Rnobc) influences whether the group produces over 100 # # verbal backchannels towards one another or not (produced_more_than_100_bcs) # # - we will use a chi square test of independence to test this relationship # # ------------------------------------------------------------------------------------------------# # remove the NA data in our dependent variable jibo_is_data_prod_more_100_bcs <- jibo_is_data[!is.na(jibo_is_data["produced_more_than_100_bcs"]), ] # create the chi square table table_bc_robot_and_prod_100_bcs <- table( jibo_is_data_prod_more_100_bcs$condition_1_Rbc_0_Rnobc, jibo_is_data_prod_more_100_bcs$produced_more_than_100_bcs ) # you can take a look at the table table_bc_robot_and_prod_100_bcs # run the test chisq.test( table_bc_robot_and_prod_100_bcs, correct=FALSE ) # ------------------------------------------------------------------------------------------------# # one categorical IV and one continuous DV --> t-test # # - we will test whether the average group psychological safety scores # # (avg_psychological_safety) are different between groups with a backchanneling robot and # # groups without a backchanneling robot (condition_1_Rbc_0_Rnobc) # # - we will use an independent samples t-test for 2 groups # # - info on other types of t-test can be found: https://www.statmethods.net/stats/ttest.html # # ------------------------------------------------------------------------------------------------# # run the t-test t.test(jibo_is_data$avg_psychological_safety ~ jibo_is_data$condition_1_Rbc_0_Rnobc) # we see that the condition of whether the robot backchannels or not does not have a significant # influence on the group's average psychological safety # ------------------------------------------------------------------------------------------------# # one categorical IV and one continuous DV --> 1-way ANOVA # # - we will test whether the average group psychological safety scores # # (avg_psychological_safety) are different between groups with a backchanneling robot and # # groups without a backchanneling robot (condition_1_Rbc_0_Rnobc) controlling for # # various factors (e.g. age, emotional intelligence, number of females) # # - we will use a 1-way ANOVA test # # - more info on the 1-way ANOVA can be found here: # # http://www.sthda.com/english/wiki/one-way-anova-test-in-r # # ------------------------------------------------------------------------------------------------# # run the 1-way ANOVA psych_safety_bc_robot_aov <- aov( avg_psychological_safety ~ condition_1_Rbc_0_Rnobc + avg_age + num_females + avg_extraversion + avg_emotional_intelligence, data = jibo_is_data) # view the ANOVA's results summary(psych_safety_bc_robot_aov) # we see that the psychological safety scores are not significantly different between the # conditions where the robot does and does not backchannel # if you're interested in seeing the mean values for the psychological safety scores for variables # that are significantly correlated with the average psychological safety scores use the aggregate # command; here we see that the psychological safety scores increase as the number of females on # the team increase aggregate( jibo_is_data$avg_psychological_safety, jibo_is_data['num_females'], mean ) # ------------------------------------------------------------------------------------------------# # two categorical IVs and one categorical DV --> logistic regression # # - we will test whether the between subjects condition of whether the robot is backchanneling # # or not (condition_1_Rbc_0_Rnobc) and the between subjects condition of whether # # the robot liaison is an insider group member or outsider group member # # (condition_2_Lin_3_Lout) influences whether the group produces over 100 # # verbal backchannels towards one another or not (produced_more_than_100_bcs), controlling # # for average extraversion, average emotional intelligence, and the number of females on the # # team # # - we will use a logistic regression (glm with binomial family with logit link) # # - you can find more out about R's glm at # # https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/glm # # ------------------------------------------------------------------------------------------------# prod_100_bcs_glm <- glm( produced_more_than_100_bcs ~ condition_1_Rbc_0_Rnobc + condition_2_Lin_3_Lout + condition_1_Rbc_0_Rnobc * condition_2_Lin_3_Lout + # the interaction term avg_extraversion + avg_emotional_intelligence + num_females, data = jibo_is_data, family = binomial, # more info on families can be found here: https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/family na.action = na.omit # this takes care of removing all of the NA data ) summary(prod_100_bcs_glm) # ------------------------------------------------------------------------------------------------# # two categorical IVs and one continous DV --> 2-way ANOVA # # - we will test whether the average group psychological safety scores # # (avg_psychological_safety) are different between groups with a backchanneling robot and # # groups without a backchanneling robot (condition_1_Rbc_0_Rnobc) as well as whether the # # robot liaison is an ingroup or outgroup member (condition_2_Lin_3_Lout), controlling for # # various factors (e.g. age, emotional intelligence, number of females) # # - we will use a 2-way ANOVA test # # ------------------------------------------------------------------------------------------------# # run the 2-way ANOVA psych_safety_2_way_aov <- aov( avg_psychological_safety ~ condition_1_Rbc_0_Rnobc + condition_2_Lin_3_Lout + condition_1_Rbc_0_Rnobc * condition_2_Lin_3_Lout + avg_age + num_females + avg_extraversion + avg_emotional_intelligence, data = jibo_is_data) # view the ANOVA's results summary(psych_safety_2_way_aov) # if you were to find a significant interaction term in the anova (p value of the interaction term # that is smaller than 0.05), then you would run pairwise comparisons to investigate that # interaction further TukeyHSD(psych_safety_2_way_aov, "condition_1_Rbc_0_Rnobc", ordered = FALSE) # if you're using a different method for performing pairwise comparisons, you may need to use some # kind of statistical correction, a common one to perform is the Bonferroni correction - which is # one of the strictest forms of corrections # if you're interested in using the Bonferroni correction, information can be found here on how to # run it: https://stat.ethz.ch/R-manual/R-devel/library/stats/html/p.adjust.html # ------------------------------------------------------------------------------------------------# # one continuous IV and one categorical DV --> Pearson correlation # # - we want to test if there is a significant linear correlation between the group's average # # emotional intelligence (avg_emotional_intelligence) and whether or not the group produces # # more than 100 backchannels (produced_more_than_100_bcs_numeric) # # - use a Pearson correlation # # - if you want to control for other factors/control variables, run a logistic regression # # instead (see logistic regression example above) # # ------------------------------------------------------------------------------------------------# # you can't run Pearson correlations on factored variables, so we'll turn our factor variable into # a numeric variable jibo_is_data$produced_more_than_100_bcs_numeric <- as.numeric(jibo_is_data$produced_more_than_100_bcs) cor.test( jibo_is_data$avg_emotional_intelligence, jibo_is_data$produced_more_than_100_bcs_numeric, method="pearson", # specify that you want it to be a Pearson correlation na.action=na.omit # omit the missing (NA) data if any ) # ------------------------------------------------------------------------------------------------# # one continuous IV and one continuous DV --> linear regession # # - we want to test whether the number of robot backchannels (num_robot_bcs) influenced the # # group's average psychological safety score (avg_psychological_safety), controlling for our # # experimental conditions as well as other factors (e.g. age, emotional intelligence, number # # of females) # # - we will use a linear regression # # ------------------------------------------------------------------------------------------------# rbc_on_ps <- lm( avg_psychological_safety ~ num_robot_bcs + condition_1_Rbc_0_Rnobc + condition_2_Lin_3_Lout + condition_1_Rbc_0_Rnobc * condition_2_Lin_3_Lout + avg_age + num_females + avg_extraversion + avg_emotional_intelligence, data = jibo_is_data ) summary(rbc_on_ps)