This repository contains just one script used to analyse changes in transport behaviour over time. The data were obtained from the Multinational Time Use Study (MTUS), part of the Integrated Public Use Microdata Series (IPUMS). MTUS data are jointly hosted by the Minnesota Population Center at the University of Minnesota, and the Centre for Time Use Research at the University of Oxford.
Data downloads are accompanied by a corresponding "ddi" file, included here both in .xml and plain text formats, as well as an equivalent .html document for easy reading. This file codifies the hierarchical structure of the data, including descriptions of all variable encodings. The data download can be reproduced by following the structure of these files.
The corresponding data are very large, and therefore not directly included within this repository. They include the following variables and corresponding codes (where "MAIN" describes the main activity):
| variable | code | description |
|---|---|---|
| MAIN | 043 | walking |
| MAIN | 044 | cycling |
| MAIN | 063 | travel to/from work |
| MAIN | 064 | education travel |
| MAIN | 065 | voluntary/civic/religious travel |
| MAIN | 066 | child/adult care travel |
| MAIN | 067 | shop, person/child care travel |
| MAIN | 068 | other travel |
| MTRAV | 01 | car |
| MTRAV | 02 | public transport |
| MTRAV | 03 | walk/on foot |
| MTRAV | 04 | other physical transport |
| MTRAV | 05 | other/unspecified |
IPUMS data can now be read directly into R using the fabulous ipumsr package. The primary functions simply need the "ddi" file describing the data structure. It is presumed here that the data themselves reside in the same directory (generally in .gz form).
library (ipumsr)
dat <- read_ipums_micro_list (ddi = "./data/mtus_00001.xml",
vars = c ("COUNTRY", "YEAR", "AGE", "SEX",
"TIME", "MAIN", "MTRAV"))
#> Use of data from MTUS-X is subject to conditions including that users should
#> cite the data appropriately. Use command `ipums_conditions()` for more details.
#>
#> Reading data...
#> Parsing data...
dat
#> $PERSON
#> # A tibble: 419,686 x 5
#> RECTYPE COUNTRY YEAR AGE SEX
#> <dbl+lbl> <chr+lbl> <int> <dbl> <int+lbl>
#> 1 2 AT 1992 41. 1
#> 2 2 AT 1992 40. 2
#> 3 2 AT 1992 70. 1
#> 4 2 AT 1992 68. 2
#> 5 2 AT 1992 78. 1
#> 6 2 AT 1992 71. 2
#> 7 2 AT 1992 74. 1
#> 8 2 AT 1992 73. 2
#> 9 2 AT 1992 58. 2
#> 10 2 AT 1992 29. 2
#> # ... with 419,676 more rows
#>
#> $ACTIVITY
#> # A tibble: 9,224,050 x 6
#> RECTYPE COUNTRY YEAR TIME MAIN MTRAV
#> <dbl+lbl> <chr+lbl> <int> <dbl> <int+lbl> <int+lbl>
#> 1 3 AT 1992 390. 2 -7
#> 2 3 AT 1992 15. 4 -7
#> 3 3 AT 1992 15. 66 1
#> 4 3 AT 1992 15. 6 -7
#> 5 3 AT 1992 15. 6 -7
#> 6 3 AT 1992 45. 6 -7
#> 7 3 AT 1992 30. 19 -7
#> 8 3 AT 1992 15. 25 5
#> 9 3 AT 1992 30. 24 -7
#> 10 3 AT 1992 15. 21 5
#> # ... with 9,224,040 more rowsNote that the dat$ACTIVITY$TIME variable quantifies the duration of a travel activity, and is the key variable we will be analysing here. These data include over 9 million activities from nearly 500,000 individuals from these countries:
countries <- unique (dat$PERSON$COUNTRY)
countries
#> [1] "AT" "CA" "FI" "FR" "NL" "ES" "UK" "US"The following analyses will use tidyverse functionality, so before we start we load all packages:
require (tidyverse)The two tables represent the hierarchical structure of the data. These data can then be filtered down to only those records describing transport activities. We will only use the ACTIVITY table here.
act <- filter (dat$ACTIVITY, MTRAV > 0 | MAIN %in% c (43:44, 63:68))
act
#> # A tibble: 1,475,679 x 6
#> RECTYPE COUNTRY YEAR TIME MAIN MTRAV
#> <dbl+lbl> <chr+lbl> <int> <dbl> <int+lbl> <int+lbl>
#> 1 3 AT 1992 15. 66 1
#> 2 3 AT 1992 15. 25 5
#> 3 3 AT 1992 15. 21 5
#> 4 3 AT 1992 15. 23 5
#> 5 3 AT 1992 135. 59 5
#> 6 3 AT 1992 15. 4 5
#> 7 3 AT 1992 30. 68 1
#> 8 3 AT 1992 15. 63 1
#> 9 3 AT 1992 15. 63 3
#> 10 3 AT 1992 15. 63 5
#> # ... with 1,475,669 more rowsWe still have almost 1.5 million travel activities. Now let's make equivalent tables for each activity, plus one for all travel activities combined (represented by MTRAV > 0):
bike <- filter (act, MAIN == 44 | MTRAV == 4)
walk <- filter (act, MAIN == 43 | MTRAV == 3)
allmodes <- filter (act, MTRAV > 0)Note that MTRAV == 4 describes "other physical transport," which may or may not represent bicycle-borne transport.
The data in the resultant tables then need to be aggregated to generate total durations of each activity for each year and country. The following function converts those durations to proportions of cycling and walking as a fraction of total travel time.
gettimes <- function (ci = "CA")
{
tb <- bike %>%
dplyr::filter (COUNTRY == ci) %>%
dplyr::group_by (YEAR) %>%
dplyr::summarize (time = sum (TIME))
tw <- walk %>%
dplyr::filter (COUNTRY == ci) %>%
dplyr::group_by (YEAR) %>%
dplyr::summarize (time = sum (TIME))
ta <- allmodes %>%
dplyr::filter (COUNTRY == ci) %>%
dplyr::group_by (YEAR) %>%
dplyr::summarize (time = sum (TIME))
if (nrow (tb) == 0)
message (ci, ": no bike data")
if (nrow (tw) == 0)
message (ci, ": no walking data")
year <- sort (unique (c (tb$YEAR, tw$YEAR, ta$YEAR)))
year <- year [which (year %in% ta$YEAR)]
times <- tibble (year = year,
walk = tw$time [match (year, tw$YEAR)],
bike = tb$time [match (year, tb$YEAR)],
allmodes = ta$time [match (year, ta$YEAR)])
times [is.na (times)] <- 0
times$walk <- times$walk / times$allmodes
times$bike <- times$bike / times$allmodes
if (nrow (times) < 3) # require >= 3 years of data
message (ci, ": no multi-year data")
return (times)
}
times <- lapply (countries, gettimes)
#> AT: no multi-year data
#> CA: no multi-year data
#> FI: no multi-year data
#> FR: no multi-year data
#> ES: no multi-year data
names (times) <- countriesMost countries do not unfortunately have sufficient data to analyse further, leaving only the UK, the US, and the Netherlands.
times <- times [which (countries %in% c ("UK", "US", "NL"))]
times
#> $NL
#> # A tibble: 5 x 4
#> year walk bike allmodes
#> <int> <dbl> <dbl> <dbl>
#> 1 1975 0. 0.345 560235.
#> 2 1985 0.116 0.230 1533420.
#> 3 1990 0.111 0.233 1701870.
#> 4 2000 0.0941 0.219 952215.
#> 5 2005 0.103 0.246 1339395.
#>
#> $UK
#> # A tibble: 8 x 4
#> year walk bike allmodes
#> <int> <dbl> <dbl> <dbl>
#> 1 1974 0.136 0. 595280.
#> 2 1975 0.0755 0. 748440.
#> 3 1983 0.162 0.00258 395955.
#> 4 1984 0.142 0.00384 304575.
#> 5 1987 0.240 0.0140 1032150.
#> 6 2000 0.354 0.0263 1065800.
#> 7 2001 0.359 0.0293 1269290.
#> 8 2005 0.206 0.0139 418660.
#>
#> $US
#> # A tibble: 18 x 4
#> year walk bike allmodes
#> <int> <dbl> <dbl> <dbl>
#> 1 1985 0.0308 0.00894 317374.
#> 2 1992 0.0739 0.00470 67275.
#> 3 1993 0.0581 0.00548 425323.
#> 4 1994 0.0596 0.00713 364896.
#> 5 1995 0.0509 0.00113 66477.
#> 6 1998 0.0306 0. 118011.
#> 7 1999 0.0332 0. 65102.
#> 8 2000 0.0255 0. 75960.
#> 9 2003 0.0617 0.00646 1893502.
#> 10 2004 0.0588 0.00679 1186800.
#> 11 2005 0.0611 0.00800 1132460.
#> 12 2006 0.0651 0.00855 1116563.
#> 13 2007 0.0645 0.00795 1039728.
#> 14 2008 0.0697 0.00967 1065144.
#> 15 2009 0.0697 0.00907 1119126.
#> 16 2010 0.0701 0.00838 1178886.
#> 17 2011 0.0664 0.00965 1082362.
#> 18 2012 0.0685 0.0108 1060431.The US also only has usable data after 1998, and the Netherlands after 1975, so
times$US <- filter (times$US, year > 1998)
times$NL <- filter (times$NL, year > 1975)These data can now be converted to a single data.frame to be plotted
times$UK$COUNTRY <- "UK"
times$US$COUNTRY <- "US"
times$NL$COUNTRY <- "NL"
times <- do.call (rbind, times)Bicycle journeys in the US and UK generally represent around one fifth of the proportion of walking journeys, so we multiply them by 5 to enable both to be plotted on the same scale
times$bike [times$COUNTRY != "NL"] <- times$bike [times$COUNTRY != "NL"] * 5Now merge the variables into a single country_trans_mode variable for easy plotting:
times <- gather (times, key = trans_mode, value = time, walk, bike)
times$country_mode <- paste0 (times$COUNTRY, ":", times$trans_mode)These data may now be used to plot changes in rates of walking and cycling over time, using the solarized colours from the ggthemes package.
require (ggthemes)
ggplot (times, aes (year, time, colour = country_mode)) +
geom_line () +
geom_point () +
theme_solarized ()
Remember that rates of bicycle trips are multiplied by 5 for the UK and US, but not for the NL. The drastic decreases in the final points for the UK (2005) seem anomalous, because it is unlikely that walking actually decreased by almost one half. If we consider all other results to be reliable, they suggest that patterns in the Netherlands have remained largely stable over 20 years, with a slight decrease in rates of walking, and a notable increase in cycling between 2000 and 2005. Active transport generally decreased in the US until 2000, after which both walking and cycling have progressively increased to respectively represent around 7 and just over 1%. Rates in the UK appear, in contrast, to have increased much more rapidly.
The following code quantifies the relative rates of change in the US post-2000, and in the UK up to and including 2001. The analysis will be grouped by COUNTRY and trans_mode, requiring the following lines to prepare the data for analysis:
timesf <- filter (times, !((country_mode == "UK:walk" & year == 2005) |
(country_mode == "UK:bike" & year == 2005) |
(country_mode == "US:walk" & year <= 2000) |
(country_mode == "US:bike" & year <= 2000))) %>%
group_by (COUNTRY, trans_mode) %>%
nest ()The analysis can then be implemented with the following code:
mod <- function (df)
summary (lm (time ~ year, data = df))
timesf <- timesf %>%
mutate (mod = map (timesf$data, mod)) %>%
mutate (coeffs = mod %>% map (broom::tidy)) %>%
unnest (coeffs) %>%
filter (term == "year")The second mutate call maps the lm models to data.frame objects of coefficients, which generates a term column containing values for both (Intercept), and year, where the last line filters only the latter. Finally, converts rates of change to annual percentages, and remove a couple of statistics to help us concentrate on what we want to know. Rates of change for cycling are also divided by 5 to account for the previous multiplication that was used for plotting.
timesf$std.error <- timesf$statistic <- NULL
timesf$estimate <- timesf$estimate * 100
timesf$estimate [timesf$COUNTRY != "NL" & timesf$trans_mode == "bike"] <-
timesf$estimate [timesf$COUNTRY != "NL" & timesf$trans_mode == "bike"] / 5
timesf
#> # A tibble: 6 x 5
#> COUNTRY trans_mode term estimate p.value
#> <chr> <chr> <chr> <dbl> <dbl>
#> 1 NL walk year -0.0869 0.179
#> 2 UK walk year 0.989 0.000582
#> 3 US walk year 0.108 0.00446
#> 4 NL bike year 0.0394 0.672
#> 5 UK bike year 0.112 0.000388
#> 6 US bike year 0.0387 0.000644All changes are significant except for those in the Netherlands. Rates of walking increased by almost 1% per year in the UK, and just over 0.1% in the US. In contrast to these changes in rates of walking, rates of cycling increased relatively much more rapidly in the US (0.04% per year) than in the UK (0.11% per year).