Advanced topics

Subsetting a survey

Consider this example, in which we estimate the number of medications by age group:

library(surveytable)
set_survey(namcs2019sv)
Survey info {NAMCS 2019 PUF}
Variables Observations Design
33 8,250 Stratified 1 - level Cluster Sampling design (with replacement)
With (398) clusters.
namcs2019sv = survey::svydesign(ids = ~CPSUM, strata = ~CSTRATM, weights = ~PATWT
, data = namcs2019sv_df)
tab_subset("NUMMED", "AGER")
Number of medications coded (for different levels of Patient age recode) {NAMCS 2019 PUF}
Level % known Mean SEM SD
Under 15 years 100 1.58 0.168 1.75
15-24 years 100 1.64 0.112 1.7 
25-44 years 100 2.15 0.225 2.74
45-64 years 100 3.49 0.303 4.49
65-74 years 100 4.44 0.431 5.03
75 years and over 100 5.53 0.494 5.59

What if we’d like to estimate the same thing, but only for the visits for which NUMMED > 0?

One way to do this is to create another survey object for which NUMMED > 0, and then analyze this new survey object.

newsurvey = survey_subset(namcs2019sv, NUMMED > 0
  , label = "NAMCS 2019 PUF: NUMMED 1+")
set_survey(newsurvey)
Survey info {NAMCS 2019 PUF: NUMMED 1+}
Variables Observations Design
33 5,738 Stratified 1 - level Cluster Sampling design (with replacement)
With (374) clusters.
survey_subset(namcs2019sv, NUMMED > 0, label = “NAMCS 2019 PUF: NUMMED 1+”)

Note that we called set_survey(), to let R know that we now want to analyze the new object newsurvey, not namcs2019sv.

Now, let’s create the table:

tab_subset("NUMMED", "AGER")
Number of medications coded (for different levels of Patient age recode) {NAMCS 2019 PUF: NUMMED 1+}
Level % known Mean SEM SD
Under 15 years 100 2.34 0.157 1.66
15-24 years 100 2.34 0.116 1.58
25-44 years 100 3.04 0.257 2.81
45-64 years 100 4.92 0.358 4.62
65-74 years 100 6.02 0.445 4.98
75 years and over 100 7.29 0.457 5.32

Be sure to check the table title to verify that you are tabulating the new survey object.

Advanced variable editing and data flow

Advanced variable editing

First, let’s review what I call “advanced variable editing”.

  1. Create a new variable in the variables data frame (which is part of the survey object).
  2. Call set_survey() again. Any time you modify the variables data frame, call set_survey().
  3. Tabulate the new variable.

For an example of this, see vignette("Example-Residential-Care-Community-Services-User-NSLTCP-RCC-SU-report").

Data flow

The above explanation raises the question of why set_survey() must be called again, after variables is modified. Here is an explanation:

The survey that you’re analyzing actually exists in three separate places:

  1. A file on your computer data storage that contains the survey object. For example, it could be an RDS file on your hard disk drive that contains the survey object named something like mysurvey.rds.
  2. The survey object in R’s global environment, named something like mysurvey.
  3. A hidden copy of the survey object that’s used by surveytable. This is what surveytable analyzes.

Why is there (3) that’s different from (2), you might ask. That’s due to an arcane issue with how R packages work – both (2) and (3) are necessary.

Normally, the information only flows forwards, from (1) to (2) and from (2) to (3).

Forwards flow:

Backwards flow:

The functions for modifying or creating variables that are part of the surveytable package (like var_cut() or var_collapse()) modify (3). Since (3) is what surveytable works with and tabulates, you can call var_collapse(), and then you can call tab(). You don’t need to do anything extra in between.

If you are modifying the variables data frame directly, you are actually modifying (2). After you modify (2), you need to copy it over to (3), so that surveytable can use it. You do that by calling set_survey().

Thus, any time you modify variables yourself, call set_survey(). You modify (2), then copy (2) -> (3) by calling set_survey().

On the flip side, the changes that you make in (3) (using surveytable functions like var_cut() or var_collapse()) are not reflected in (2). If you make changes in (3), then call set_survey(), those changes are lost, because set_survey() copies (2) -> (3). If those changes were important, you can just rerun the code that created them. If you really need to go from (3) to (2), use mysurvey = surveytable:::.load_survey().