library(lubridate)

# first create a text vector with the dates
datez <- paste(raw_data$year, raw_data$month,raw_data$day, sep = "-" )

# then create a vector with the hour and minutes
hour_min <- paste( raw_data$hour, raw_data$minute,sep = ":" )

# make one variable with all information
# and turn it into a date object with lubridate
raw_data$date_time <- ymd_hm(paste(datez,hour_min, sep = " "))

# get the last observation per day
# this line is what the csv spits out
last_obs_day <- raw_data[!duplicated(raw_data[,c('year','month','day')], fromLast=T),]

# get the observations that are recorded every half-hour
# this appears to be the standard- every day at peak hours
# data are written out on the half-hour
data_0_30 <- raw_data[raw_data$minute == 0 | raw_data$minute == 30 , ]

# put the two dataframes together
walking <- rbind(data_0_30,last_obs_day)

# drop duplicates on the date_time variable
# because some of the last observations might also just have
# been written at minutes 0 or 30
walking_omnibus <- walking[!duplicated(walking$date_time, fromLast=T),]

# order by the date object created with lubridate above
# dataset should now contain observations written out
# on the half-hour, along with the last observation for the day
# if it was not on the half-hour
library(plyr); library(dplyr)
walking_omnibus <-arrange(walking_omnibus, date_time)

# add a day break variable and extract the date of the week (using lubridate package)
walking_omnibus$oneday <- cut(walking_omnibus$date_time, breaks = "1 day")
walking_omnibus$dow <- wday(walking_omnibus$date_time, label = TRUE)

# aggregate by hour 
# because the step count is cumulative,
# we need to take the maximum value within
# a given day/hour time frame
aggregate_day_hour <- walking_omnibus %>% group_by(oneday, dow, hour) %>% 
  summarise(steps = max(steps))

I then plotted the data (which reduces to 12,958 observations- quite a lot still) using ggplot2. Because the trend is difficult to see from the points, I’ve added a loess smoother, which gives the expected step count at any given hour of the day. For example, it looks like on average I reach the 10,000 step mark at around 3 PM. 

# load the ggplot2 package
library(ggplot2)

# plot out the max counts per hour
# loess gives expected average at a given hour
p = ggplot(data = aggregate_day_hour, aes(x = hour, y = steps)) + 
  geom_point() +
  coord_cartesian(ylim = c(0, 20000)) +
  geom_smooth(method="loess", fill=NA) +
  theme(legend.title=element_blank()) +
  labs(x = "Hour of Day", y = "Steps" )
p

I then plotted out the trend lines for each day separately, to see if the evolution of step count is different on the different days of the week:

# different lines per day of week
# Sunday is the day that has the lowest step count
# Tuesday appears to be the highest
p = ggplot(data = aggregate_day_hour, aes(x = hour, y = steps, color = dow)) + 
  geom_point() +
  coord_cartesian(ylim = c(0, 20000)) +
  geom_smooth(method="loess", fill=NA) +
  theme(legend.title=element_blank()) +
  labs(x = "Hour of Day", y = "Steps" )
p

Indeed it is! The weekdays (Monday through Friday) are similar to each other, and I appear to walk more on these days. Saturday and especially Sunday are lower in terms of the step counts for basically the entire day- also Saturday and Sunday start to diverge from each other around 11 AM or so and the difference between them grows over the course of the day.

This makes sense to me- I walk to work during the weekday, and it’s no surprise that the step counts for those days are pretty much the same. I have an evening activity on Tuesday nights, and so it doesn’t surprise me that on Tuesday evenings I walk more. I do try to be active during the weekend, but it also doesn’t surprise me that (especially on Sundays) I walk less during the other days of the week.

There are 7 different lines plotted on the above graph, which is nice but is perhaps a bit too much information for a single plot. Below, I add a “weekday vs. weekend” variable to the aggregated dataset and plot out the trends based on this variable. It preserves the essential insight provided in the above plot, but is easier to read and interpret.

# add a weekday/weekend variable to the aggregated dataset
aggregate_day_hour$week_weekend[aggregate_day_hour$dow == 'Sun'] <- 'Weekend'
aggregate_day_hour$week_weekend[aggregate_day_hour$dow == 'Sat'] <- 'Weekend'
aggregate_day_hour$week_weekend[aggregate_day_hour$dow == 'Mon'] <- 'Weekday'
aggregate_day_hour$week_weekend[aggregate_day_hour$dow == 'Tues'] <- 'Weekday'
aggregate_day_hour$week_weekend[aggregate_day_hour$dow == 'Wed'] <- 'Weekday'
aggregate_day_hour$week_weekend[aggregate_day_hour$dow == 'Thurs'] <- 'Weekday'
aggregate_day_hour$week_weekend[aggregate_day_hour$dow == 'Fri'] <- 'Weekday'

# plot out steps per hour by week/weekend
# even by 8 AM on a weekend I'm behind my weekday step count
# the difference continues to grow throughout the day
p = ggplot(data = aggregate_day_hour, aes(x = hour, y = steps, color = week_weekend)) + 
  geom_point() +
  coord_cartesian(ylim = c(0, 20000)) +
  geom_smooth(method="loess", fill=NA) +
  theme(legend.title=element_blank()) +
  labs(x = "Hour of Day", y = "Steps" )
p

The pattern between weekday and weekend is quite clear. On the weekend, even by 8 AM I’m below my weekday average, and the difference between the weekday vs. weekend continues to grow throughout the day. On a weekday, I hit 10,000 steps around 2:30 PM, but on the weekend I don’t hit 10,000 steps until around 4:30 PM (on average). By midnight, it looks like there’s more than a 3,000 step difference between the weekdays and the weekend.

That’s all for this post. I have been doing some other analyses of these data, and will present that here when I have the time!