Retrieve and summarize data • psrc.travelsurvey

Works with travelSurveyTools

The psrc.travelsurvey package is meant to complement and enhance the travelSurveyTools package for PSRC users. It delivers data in the hts_data format appropriate as input to subsequent travelSurveyTools functions. It also provides wrapper functions that provide PSRC defaults to travelSurveyTools functions, resulting in simpler workflows with fewer required arguments. psrc.travelsurvey is constrained to an extent by fitting into the travelSurveyTools system, but if you have suggestions for other features or functions you’d find helpful, please contact the authors and we’ll see if we can implement them.

Data retrieval

Since psrc.travelsurvey uses Elmer, the agency’s central database, to retrieve PSRC household travel survey data you’ll need to be connected (in office or via VPN). Then use get_psrc_hts(), which takes two arguments:

survey_years - a vector of one or more years; the default is all available i.e., c(2017,2019,2021,2023) as of this writing.
survey_vars - a vector of variable names (regardless of the table in which they occur)

The return object is a list with separate data.table elements for each potential unit of analysis, i.e. household (hh), person, day, trip, and vehicle. The travelSurveyTools package workflow uses this central object for related or independent variable summaries. It can include the full set of variables for all your summaries, so it’s unnecessary to request separate data objects to feed each separate summary data table. It also supports multiyear summaries (trends/comparisons) via the survey_year field.

For efficiency, get_psrc_hts() only requests the variables you specify, so you must know their precise names. A lookup function, psrc_hts_varsearch() can assist; it takes a single string argument as the codebook search term and matches either the variable name or description fields. Regular expressions are supported for more complex searches; for example, the word boundary operator \\b can be used to omit vehicle model from results when searching for “mode”.

View(psrc_hts_varsearch("\\bmode\\b"))

To browse the entire variable list, open the internal data object init_variable_list:

View(psrc.travelsurvey:::init_variable_list)

Note: It is unnecessary to include the survey year, table ids, weights, or survey stratification variable (sample_segment) in your survey_vars argument; since they are necessary for later steps, get_psrc_hts() includes them implicitly. Factor variables and levels are applied from the codebook, and missing or skip-value codes are recoded as NA.

Summarization

To summarize data, use psrc_hts_stat(). This has four primary arguments, and one option. The arguments are:

hts_data - the hts_data object
analysis_unit - string specifying the scale at which statistics are calculated, either “hh”, “person”, “day”,“trip”, or “vehicle”.
group_vars - any grouping variables, in nesting order (share summaries are calculated per categories of the variable listed last).
stat_var (optional) - Numeric variable if min/max/median/mean is desired; omit for count/share summaries.

As an example, we’ll retrieve the following survey variables, then do one count/share summary, and one numeric summary. Notice, travelSurveyTools handles the relational element of the data, so tables lower in the organizational heirarchy (like trips) inherit attributes from those above them (such as hh); this is a handy feature and avoids unnecessary data duplication among tables.

library(psrc.travelsurvey)
library(data.table)

# Specify which variables to retrieve
vars <- c("hhsize", "employment", "commute_mode", "dest_purpose",
          "mode_class_5", "distance_miles")

# Retrieve the data
hts_data <- get_psrc_hts(survey_vars = vars)  # default includes all survey_years
  
# Calculate a numeric summary, i.e. min/max/median/mean
rs1 <- psrc_hts_stat(hts_data, 
                     analysis_unit="trip", 
                     group_vars="hhsize", 
                     stat_var="distance_miles")

head(dplyr::select(rs1[survey_year==2023], -c(survey_year, min, max)))
#>      hhsize count      mean   mean_moe   median
#>       <ord> <int>     <num>      <num>    <num>
#> 1: 1 person 10081  6.977442  1.4370516 2.222023
#> 2: 2 people 12719 11.073412  3.2645435 3.135158
#> 3: 3 people  3915  7.009381  0.7154436 3.035398
#> 4: 4 people  3858  6.123352  0.5365826 3.197576
#> 5: 5 people  1152 23.016527 12.1648649 2.116390
#> 6: 6 people   404  5.836721  1.1624158 2.054875

# Calculate a categorical summary, i.e. count/share
rs2 <- psrc_hts_stat(hts_data, "trip", c("hhsize", "mode_class_5"))

head(dplyr::select(rs2[survey_year==2023 & as.character(mode_class_5)=="Transit"],
                   -c(survey_year, mode_class_5)))
#>      hhsize count       prop    prop_moe       est  est_moe
#>       <ord> <int>      <num>       <num>     <num>    <num>
#> 1: 1 person   850 0.07819693 0.014343472 144804.30 27878.08
#> 2: 2 people   553 0.02450678 0.005668314 121841.55 28456.42
#> 3: 3 people   124 0.01886768 0.006262066  55321.98 18366.04
#> 4: 4 people    80 0.01031612 0.003696247  37591.82 13442.13
#> 5: 5 people    28 0.02357459 0.008536841  35762.24 12806.51
#> 6: 6 people    14 0.05805984 0.028137362  57373.12 28499.56

Custom variables and filtering

Since each hts_data element is a data.table, they can be manipulated with any combination of base r, tidyverse (e.g. mutate()), or data.table (e.g :=) syntax. The example below illustrates adding variables specific to the hts_data$trip element. Notice, the inheritance capability mentioned earlier applies to custom variables as well, so if you want to recode a household-level variable to feature in a trip-level analysis, just add it to hts_data$hh (not to hts_data$trip).

Standard category groupings not present in the base data can be added via the functions described in the accompanying vignette. These take the entire hts_data object as argument, not a specific table (the package knows which is involved). This lends itself to piping multiple standard recodes.

Keep in mind that filtering any table removes records available for subsequent analyses. Rather than manage multiple filtered copies of the data object, you may want to create a new variable on the relevant table with NA values for the records you intend to exclude from the summary. You can then use the incl_na=FALSE option to exclude that line from summary results (of particular value when you want shares calculated from only reported categories). Here we calculate mode split percentages for work trips only by adding a variable coded NA for all non-work-related trips, then using the incl_na=FALSE option to limit statistics to that category.

library(dplyr)
library(stringr)

# Use a standard variable recode to get simpler trip purpose
hts_data <- hts_bin_dest_purpose(hts_data)

# Use dplyr::mutate to add a simplified mode field & work purpose specifier
hts_data$trip <- mutate(
  hts_data$trip, 
  purpose_work = case_when(
    dest_purpose_bin4=="Work" ~ "Work",
    !is.na(dest_purpose_bin4) ~ NA_character_)
) 

# Calculate a categorical summary, i.e. count/share
rs3 <- psrc_hts_stat(hts_data, "trip", c("purpose_work", "mode_class_5"), incl_na=FALSE)

head(dplyr::select(rs3[survey_year==2023],
                   -c(survey_year, purpose_work)))
#>    mode_class_5 count        prop    prop_moe        est    est_moe
#>           <ord> <int>       <num>       <num>      <num>      <num>
#> 1:        Drive  3211 0.865739816 0.018028620 1835881.03 125651.228
#> 2:      Transit   402 0.029027564 0.007653030   61555.62  16093.543
#> 3:         Walk   728 0.079432698 0.013730924  168444.35  29070.846
#> 4:         Bike   153 0.008963787 0.002837518   19008.54   5922.878
#> 5:        Other   118 0.016836136 0.008717198   35702.58  18613.062

Note: travelSurveyTools summary functions require additional preparatory steps that psrc.travelsurvey handles for you, for simplicity. To use native travelSurveyTools summary functions, you’ll need to handle these extra steps yourself, since travelSurveyTools does not have a dependency on the psrc.travelsurvey package.