Skip to contents

Retrieve and summarize data

retrieve-and-summarize-data.Rmd

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", 
          "mode_characterization", "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 10075  6.380222  0.8284701 2.169828
#> 2: 2 people 11871 13.174915  3.0272559 3.311287
#> 3: 3 people  3517  8.357221  0.9016603 4.094215
#> 4: 4 people  3595 10.254066  5.6922574 3.233616
#> 5: 5 people  1018 32.329314 19.3722674 2.761995
#> 6: 6 people   371  5.709866  0.8525478 3.046583

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

head(dplyr::select(rs2[survey_year==2023 & as.character(mode_characterization)=="Transit"],
                   -c(survey_year, mode_characterization)))
#>      hhsize count       prop    prop_moe       est   est_moe
#>       <ord> <int>      <num>       <num>     <num>     <num>
#> 1: 1 person   908 0.08090005 0.012606740 173635.72 28466.102
#> 2: 2 people   563 0.02056404 0.004167832  88135.46 17783.128
#> 3: 3 people   165 0.01946050 0.005394446  32299.75  8864.421
#> 4: 4 people   205 0.04214847 0.009085101 162495.97 35391.190
#> 5: 5 people    77 0.06091950 0.015838454 131608.22 34802.763
#> 6: 6 people    49 0.10200039 0.026777949 140783.94 38384.352

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. Additionally, standard category groupings not present in the base data can be added via the functions described in the accompanying vignette.

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 will add a dichotomous worker variable, then calculate shares with NA values excluded (they would otherwise comprise about 20 percent of the total).

# Add the worker variable 
hts_data <- hts_bin_worker(hts_data)

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

head(rs3)
#>    survey_year     worker count      prop   prop_moe       est   est_moe
#>          <int>     <fctr> <int>     <num>      <num>     <num>     <num>
#> 1:        2017     Worker  3863 0.6681291 0.03075054 1991691.5 139270.04
#> 2:        2017 Not Worker  1517 0.3318709 0.03075054  989306.5 104621.62
#> 3:        2019     Worker  3585 0.6718722 0.03146935 2024432.5 142762.94
#> 4:        2019 Not Worker  1325 0.3281278 0.03146935  988688.8 109105.54
#> 5:        2021     Worker  2125 0.6848086 0.01923162 2268045.2  88724.59
#> 6:        2021 Not Worker  1313 0.3151914 0.01923162 1043894.8  66403.29

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.