Let’s download our data from experiments in Netlogo and take a look at the results.

library(tidyverse)
data_model <- read.csv("C:/Users/Anastasia/Downloads/model_vaccinate-anti-vaccinate_5_experiment1-table.csv", skip = 6)

data_model$mech <- as.character(data_model$mech)

data_with <- data_model %>%
  filter(decision.making.on.every.tick. == "true")
data_not <- data_model %>%
  filter(decision.making.on.every.tick. == "false")

Now let’s plot the two different situation: when we switch between decision-making being done on every tick or just in the beginning of the setup.

ggplot() +
  geom_boxplot(data = data_not, aes(x = mech, y = X..count.turtles.with.opinion....anti.vaccinate.......count.turtles.....100))  +
  labs(x = 'Type of network', y = '% agents with anti-vaccibate opinion',
       title = "Distribution of percentage of anti-vaccinate opinion in different networks \n with decision-making only in the beginning") + 
  theme_bw()

Here we see the boxplots for three network types. Have a look at the medians in these - we can clearly see that in the random-attachment network the spread of an idea is incredibly high, making it 40% (median) and only 2% for preferential attachment, for instance, and 8% for small world.

ggplot() +
  geom_boxplot(data = data_with, aes(x = mech, y = X..count.turtles.with.opinion....anti.vaccinate.......count.turtles.....100))  +
  labs(x = 'Type of network', y = '% agents with anti-vaccibate opinion',
       title = "Distribution of percentage of anti-vaccinate opinion in different networks \n with decision-making on every tick") + 
  theme_bw()

Now, what happens here? Woah, the changes are remarkable. If agents can change their mind on every tick, preferential attachment network is still unaffected by the ideas of anti-vaccination, whereas both random attachment and small world networks are affected immensely, especially compared to the previous plot.

What do we get from these? We can compare the networks now.

  1. In both cases we noticed how little the first network is affected by the idea spreading. It is due to the fact that this network is constructed in such a way that agents tend to link with the most popular agents, who might no be a part of anti-vaccination movement at all. Thus, we see little spreading in preferential attachment network as it would be, for instance, in a community with a prominent leader who holds his belief and others follow him on that.

  2. When links between agents appear randomly, chances are, the idea will spread to a large share of population. Since clustering is absent and agents do not form groups based on their attributes, we witness the case of highly heterogeneous system where an idea spreads really fast.

  3. The small world network represents a society where you don’t really know every person but you can be familiar with them “through” other people. This one actually seems the most interesting since it is how an idea would spread in the community of people who might not know each other in person. We consider this one and the first one to be the most prominent and true of the real world.