Time series in grouped dataframe

Question

I am currently dealing with multiple time series within a grouped/nested dataframe and I do not know how to proceed. I hope that someone here can help me on this matter.

My dataset consist of 5 grouping variables (Category), each of which contains 10 replicates (ID). Each of this replicate consists in 60 consecutive observations at regular time intervals (Time). Each of this observations then consists of multiple variables: Amount, Walk_1, Walk_2 and Walk_1+2. However, I also have three other variables (Amount1, Amount2, Amount3) which consist in the Number variable splitted by area.

So, to have it graphically:

A { (1, 2, 3, 4, 5, 6, 7, 8, 9, 10) { 1(1, 2, 3, 4, ... , 60); 2(1, 2, 3, 4, ... , 60); ..and so on.

B { (11, 12, 13, 14, 15, 16, 17, 18, 19, 20) { 11(1, 2, 3, 4, ... , 60); 12(1, 2, 3, 4, ... , 60); ..and so on.

.

.

E { (41, 42, 43, 44, 45, 46, 47, 48, 49, 50) { 41(1, 2, 3, 4, ... , 60); 42(1, 2, 3, 4, ... , 60); ..and so on.

What I want to know is whether the variables walk_1, walk_2 and walk_1+2:

• Differ within and across the grouping variables Category or ID
• show an oscillatory/periodic trend over the 60 observations.

I tried to nest the dataset into ID categories, then apply the acf function by group using the mutate + map function. This way, I obtain a series of acf values. However, I do not know how to proceed for visualising and analysing the dataset.

Is this method functional to my purpose, or should I use a different function?

EDIT: Since I was asked to put a reproducible example, here is some similar data structure with a simplified fake dataset.

dat1 = data.frame(
category = rep(x = 1:5, each=600),
ID = rep(rep(1:50,each=60)),
walk_1 = rnorm(n = 3000,mean = 10,sd = 4),
walk_2 = rnorm(n = 3000, mean = 5, sd = 5),
amount = sample.int(50, 3000, replace = TRUE))

dat1 <- dat1 %>%
  group_by(ID) %>% 
  mutate(walk_12 = walk_1 + walk_2) %>%
  nest() 

dat1 <- dat1 %>%
  mutate(data = map(data, 
                   ~ mutate(., Time = seq(1, 60, by = 1)))) %>%
  unnest() 
  
dat1 <- dat1 %>%
   mutate(walk_1 = ifelse(Time == 1, NA, walk_1)) %>%
   mutate(walk_2 = ifelse(Time == 1, NA, walk_2)) %>%
   mutate(walk_12 = ifelse(Time == 1, NA, walk_12))

And here's the output of dput(head(dat1)):

structure(list(ID = c(1L, 1L, 1L, 1L, 1L, 1L), category = c(1L, 
1L, 1L, 1L, 1L, 1L), walk_1 = c(NA, 7.93428744179875, 11.6326574689602, 
10.3687325793843, 6.2631358473095, 14.0134490135895), walk_2 = c(NA, 
-4.03741066457775, 2.91290193315445, 8.04203547142631, 9.42608080771425, 
16.2253066800552), amount = c(34L, 37L, 31L, 26L, 29L, 33L), 
    walk_12 = c(NA, 3.896876777221, 14.5455594021146, 18.4107680508106, 
    15.6892166550237, 30.2387556936447), Time = c(1, 2, 3, 4, 
    5, 6)), row.names = c(NA, -6L), groups = structure(list(ID = 1L, 
    .rows = structure(list(1:6), ptype = integer(0), class = c("vctrs_list_of", 
    "vctrs_vctr", "list"))), row.names = 1L, class = c("tbl_df", 
"tbl", "data.frame"), .drop = TRUE), class = c("grouped_df", 
"tbl_df", "tbl", "data.frame"))

@Marco Here, amount is a count of individuals and thus previous measurements within the same ID should not be independent. Amount1, Amount2 and Amount3 are correlated in the sense that Amount1 + Amount2 + Amount3 = amount.

What I need to do is to understand whether walk_1, walk_2 and (consequently) walk_1+2 have periodicity and, if tha is the case, whether this periodicity is equal among different ID within or across the different Category. Also, I do not have data for any first time point in each ID, so the time series to analyse is 59 points.

I hope that this clarify my question. Thanks for your help!

Answer 1

For comparing statistics between categories and IDs use dplyr. The time series analysis is not trivial to do and with only 60 time points, it is not easy to find meaningful periodics. Anyway, I have good experience doing this with prophet. Please give an example on how AmountX,Y,Z is plays a role. It is unclear for me.

#example data
dat = data.frame(
category = rep(x = 1:5,each=600),
ID = rep(rep(1:10,each=60),5),
time = as.POSIXct(rep(rep(1:60,10),5)*100000,origin="1970-01-01"), # time frame 2 month + a little
walk_1 = rnorm(n = 3000,mean = 10,sd = 4)
)
##to analyse between categories or ID usw group_by from the dplyr package
dat %>% group_by(category) %>% summarise(sddev = sd(walk_1))
#or some other interesting statistics besides standard deviation

dat %>% group_by(ID) %>% summarise(sddev = sd(walk_1)) # to compare repliactes
dat %>% group_by(category, ID) %>% summarise(sddev = sd(walk_1)) # to compare repliactes and categories

#to investigate for periodic trends is not that trivial. i love to use prohpet for this https://facebook.github.io/prophet/

library(prophet)
#make a time series analysis for category 1 and replicate 1
a = dat %>% filter(category==1,ID==1) %>% mutate(ds =  time,y=walk_1) %>% select(-walk_1,-time) %>% prophet()
future <- make_future_dataframe(a, periods = 365)
b =  predict(a, future)
b$trend # the long term trend
b$weekly # weekly trend
#a plot of trends
prophet_plot_components(m, b) 
    #please note I use random data,