Introduction

Nowadays, it’s more and more popular for people to analyze data using interactive visualization. R as a powerful tool undoubtedly provides several packages to make interactive data visualization. In this report, we would like to explore and compare two functions in plotly package: ggplotly and plot_ly. Generally, ggplotly is an interactive, browser-based charting library based on ggplot2 and plot_ly is an independent graphing library makes interactive, publication-quality graphs. They have something in common while there are still some differences between these two functions.

The dataset we used comes from different sources including federal, state and local programs. It is about the affordable housing units which represent housing received financial assistance under any government program by town from Year 2011 to 2016. The dataset contains town, year, total units counted in 2010(X2010.Census.Units), units in 4 different types of assistance(Government Assisted, Tenant Rental Assistance, Single Family CHFA/USDA Mortgages and Deed Restricted), Total assisted Units and affordable housing percentage(Percent.Affordable).

housing <- read.csv("Affordable_Housing_by_Town_2011-Present.csv")
head(housing, 5)
##   Code        Town Year X2010.Census.Units Gov.Assisted
## 1    1     Andover 2016               1317           18
## 2    2     Ansonia 2016               8148          347
## 3    3     Ashford 2016               1903           32
## 4    4        Avon 2016               7389          244
## 5    5 Barkhamsted 2016               1589            0
##   Tenant.Rental.Assistance Single.Family.CHFA..USDA.Mortgages
## 1                        0                                 22
## 2                      658                                104
## 3                        2                                 33
## 4                        8                                 31
## 5                        7                                 13
##   Deed.Restricted.Units Total.Assisted.Units Percent.Affordable
## 1                     0                   40               3.04
## 2                     9                 1118              13.72
## 3                     0                   67               3.52
## 4                     0                  283               3.83
## 5                     0                   20               1.26

We basically want to analyze the percent of assisted housing and compare the four different assistance types for different year. So we used tidyverse package to group by year and gather the four assistance type in a new data frame.

housing_bar <- housing %>% group_by(Year) %>% mutate_each(funs(mean))%>%select(5:8)%>%tidyr::gather("assistance_type", "mean", 2:5) 
## Adding missing grouping variables: `Year`
head(housing_bar,5)
## # A tibble: 5 x 3
## # Groups:   Year [1]
##    Year assistance_type  mean
##   <int> <chr>           <dbl>
## 1  2016 Gov.Assisted     537.
## 2  2016 Gov.Assisted     537.
## 3  2016 Gov.Assisted     537.
## 4  2016 Gov.Assisted     537.
## 5  2016 Gov.Assisted     537.

Analyzing data

We now simply visualize and analyze this dataset to compare two functions.

1. Histogram

We focus on 2016 and want to see the overall affordable housing percentage distribution pattern. It shows that few affordable percentages are over 20, most are under 10 percent and the distribution is right skewed.

ggplotly

housing2016 <- housing[housing$Year == 2016,]
housing$Year <- as.factor(housing$Year)
g <- ggplot(housing2016,aes(x = Percent.Affordable)) + geom_histogram(bins = 20, col = "grey", fill = "lightblue", center = 1) 
ggplotly(g)

plot_ly

plot_ly(housing[housing$Year==2016,], x = ~Percent.Affordable) %>% add_histogram()

2. Boxplot

We also draw a boxplot to see the distribution, more specific on outliers. Obviouly from the boxplot, there exist a few outliers. The specific value of outliers will be shown when you click the point.

ggplotly

g <- ggplot(housing2016,aes(y = Percent.Affordable)) + geom_boxplot() 
ggplotly(g)

plot_ly

plot_ly(housing2016,y=~housing2016$Percent.Affordable,boxpoints = "suspectedoutliers") %>% add_boxplot(x = "Overall")

3. Heatmap

For the overall affordable percentage, let’s see its color distribution in different years. The affordable percentage is most gathered in the range of less than 5.

ggplotly

library(viridis)
g <- ggplot(housing,aes(x = Year, y = Percent.Affordable)) + geom_bin2d(binwidth = 3)+  scale_fill_viridis()
ggplotly(g)

plot_ly

plot_ly(x = ~housing$Year, y = ~housing$Percent.Affordable, z = ~housing$Percent.Affordable) %>% add_histogram2d(colorscale = "Blues")

4. Density

Then we dig in to specific types. We want to see the density curve for 4 different assistance types in 2016. It tells that Deed restricted units and tenant rental are gathered while government assisted units and single family CHFA/USDA Mortgages are relatively separated.

ggplotly

housing_tidy <- housing2016 %>% tidyr::gather("assistance", "numbers", 5:8)
y <- density(housing2016$Gov.Assisted)
g <- ggplot(housing_tidy,aes(x = numbers)) + geom_density(aes(fill = assistance), alpha = 0.5) + xlim(0,300)
ggplotly(g)

plot_ly

fit <- density(housing2016$Gov.Assisted)
fit1 <- density(housing2016$Tenant.Rental.Assistance)
fit2 <- density(housing2016$Single.Family.CHFA..USDA.Mortgages)
fit3 <- density(housing2016$Deed.Restricted.Units)
plot_ly(x = ~fit$x, y = ~fit$y, type = "scatter",  mode = "lines", fill = "tozeroy", yaxis = "y", name = "Gov.Assisted") %>% layout(xaxis = list(range = c(0,300))) %>% add_trace(x = ~fit1$x, y = ~fit1$y, type = "scatter",  mode = "lines", fill = "tozeroy", yaxis = "y", name = "Tenant.Rental.Assistance") %>% layout(yaxis = list(overlaying = "y", side = "right"), xaxis = list(range = c(0,300))) %>% add_trace(x = ~fit2$x, y = ~fit2$y, type = "scatter",  mode = "lines", fill = "tozeroy", yaxis = "y", name = "Single.Family.CHFA..USDA.Mortgages") %>% layout(yaxis = list(overlaying = "y", side = "right"), xaxis = list(range = c(0,300))) %>% add_trace(x = ~fit3$x, y = ~fit3$y, type = "scatter",  mode = "lines", fill = "tozeroy", yaxis = "y", name = "Deed.Restricted.Units") %>% layout(yaxis = list(overlaying = "y", side = "right"),xaxis = list(range = c(0,300))) %>% layout(legend = list(x = 0.55, y = 1.0))

5. Barplot

What about the mean of 4 types in different year? We can see that the change in different years are not so obvious, they are stable. For the mean, the government assisted units are the most and the deed restricted units are the less.

ggplotly

housing_tidy2 <- housing %>% group_by(Year) %>% mutate_each(funs(mean)) %>% select(5:8) %>%
tidyr::gather("assistance_type", "mean", 2:5) 
g <- ggplot(housing_tidy2,aes(x = assistance_type, y = mean, fill = Year)) + geom_bar(stat = "identity", position = "dodge") + scale_fill_brewer(palette = 4)+ theme(axis.text.x = element_text(size = 7, angle = 10)) 
ggplotly(g)

plot_ly

housing_bar <- housing %>% group_by(Year) %>% mutate_each(funs(mean))%>%select(5:8)%>%tidyr::gather("assistance_type", "mean", 2:5) 
plot_ly(x = ~housing_bar$assistance_type, y = ~housing_bar$mean, color = ~factor(housing_bar$Year)) %>% add_bars() %>% layout(barmode = "group")

6. Scatterplot

Finally, let’s see if there is any correlationship between 4 assistance types. It displays some correlations and the relationship between governmenr assistance and tenant rental is the closest. The differences through years are not obvious.

ggplotly

library(GGally)
g <- ggpairs(housing, columns = 5:8, aes(color = Year, alpha = 0.5))
ggplotly(g)

plot_ly

plot_ly() %>% add_trace(type='splom',dimensions=list(list(label='Gov.Assisted',values=~housing$Gov.Assisted),list(label='Tenant.Rental',values=~housing$Tenant.Rental.Assistance),list(label='Single',values=~housing$Single.Family.CHFA..USDA.Mortgages),list(label='Deed.Restricted.Units', values=~housing$Deed.Restricted.Units)),text=~class,marker = list(color = as.integer(housing_bar$Year),size = 6,line = list(width = 1)))

Conclusion

We can see that two functions are roughly the same thing: they are from the same package and they both make interactive graphs. However, plot_ly can make a lot more interactive graphs than ggplotly since it is obvious that ggplotly is simply a way to convert static plots generated from ggplot2 into interactive version while plot_ly is an independent way to directly make an interactive version of graphs.

Initially, plotly was designed as a complement for ggplot2 and plot_ly with its own unique structure is the result for that. ggplotly is a combination of plotly and ggplot2, making it easier to make interactive graphs for people who never know plot_ly but are familiar with ggplot2. So, neither ggplotly nor plot_ly is designed to replace the other. They are definitely designed for different target users.

As for the features, plot_ly has a unique and complete structure but ggplotly is dependent on ggplot2. The forms are as follows:

ggplotly(p = ggplot2::last_plot(), width = NULL, height = NULL, tooltip = “all”, dynamicTicks = FALSE, layerData = 1, originalData = TRUE, source = “A”, …)

plot_ly(data = data.frame(), …, type = NULL, name, color, colors = NULL, alpha = NULL, stroke, strokes = NULL, alpha_stroke = 1, size, sizes = c(10, 100), span, spans = c(1, 20), symbol, symbols = NULL, linetype, linetypes = NULL, split, frame, width = NULL, height = NULL, source = “A”)

Because of the different features of ggplot2 and plotly, we actually encountered problems when using plot_ly to plot density curves. Since there is not a specific function for density, we need to calculate the density at first and then plot the result points connected with line. This causes some visible differences between the density plot used ggplotly and the one used plot_ly. Also because of the fixed structure of plot_ly, the code is very tedious which need be considered carefully. And two function have different defaults, we sometimes need to take further actions to make them look exact the same, like in histograms, the default boundary for ggplot is not 0 and we need to set the center.

Besides, plot_ly provides a more direct interface to plotly.js so we can leverage more specialized chart types (such as maps) or even some visualization that the ggplot2 won’t ever support. It is better suited for more complicated tasks while the tasks ggplotly could undertake are more based on ggplot2.

The learning curves for the two tools are pretty different. It is relatively easy to learn and use ggplotly because it just need to be added after finishing graphs using ggplot2, although it has some fixed structures used to adjust things like legend and label as well. For plot_ly, it has a unique and completed structure which needs people to spend time learning, however after some learning, people can use it readily and conveniently.

For ggplotly, the pros of it are easy to learn and easy to use once people have basic knowledge of ggplot2. The cons of it are the limitation of itself which makes the truth that not all the graphs created by ggplot2 could be transferred to dynamic interactive ones.

For plot_ly, it is appropriate to almost all the dynamic interactive plots. People could use it to plot whatever they want including 3D graphs. However, it takes up a large part of memory which will make the running process very slow.

The quality of documentations of two tools is pretty similar. They both have arguments explanations and examples included which are very users friendly. Their package plotly has been developing for several versions. The latest package version on R is 4.8.0 and news can be found on this website: https://github.com/ropensci/plotly/blob/master/NEWS.md. We can see new features, improvements and bug fixed and therefore, two functions are still in active development.