Skip to contents

Compute survey means with CIs, SD, and N for grouped complex designs

Source: R/get_means.R
get_means.Rd

get_means() computes design-based means and confidence intervals for a single numeric variable from complex survey objects, optionally grouped by one or more variables. It supports survey.design, svyrep.design, srvyr::tbl_svy, and plain data.frame inputs (with a simpler weighted/unweighted implementation).

Usage

get_means(
  data,
  x,
  group = NULL,
  wt = NULL,
  decimals = 3,
  na.rm = TRUE,
  conf_level = 0.95,
  df = Inf
)

Arguments

data

A survey object (survey.design, svyrep.design, or srvyr::tbl_svy), or a plain data.frame.

x

The variable with which you wany to get the mean. Variable must be numeric but can be input as a string or symbol (e.g., "var" or var).

group

<tidy-select> A selection of columns to group the data. This operates very similarly to .by from dplyr (for more info on that see ?dplyr_by). It can also be a character vector. If using an external character vector it must be wrapped in curly brackets ({{}}).

In addition, grouped data can be piped in via dplyr::group_by() or srvyr::group_by(). If data is a grouped_df and group is provided, get_means() will combine the variable(s) used in either group_by function and the variable(s) supplied in group to calculate means.

wt

Optional weights column name (for data.frame method only). Ignored for survey objects, since weights/replicates are stored in the design.

decimals

Number of decimals to round the results to. Default is 3.

na.rm

Logical; whether to remove rows with missing values in x and group before computing frequencies. Default is TRUE.

conf_level

Determine the confidence level for Wald CIs. Default is 0.95.

df

Degrees of freedom for t critical values. Defaults to Inf (z-based). For designs, you may prefer survey::degf(design); for replicate designs, Inf is conventional.

Value

A tibble with one row per group (or one row if ungrouped). With the following columns:

  • grouping columns (preserving factor/label semantics for display)

  • mean: design-based mean of x

  • sd: weighted population standard deviation within each group

  • n: sum of weights within each group (weighted N)

  • conf_low, conf_high: Wald confidence interval bounds

The result has class "adlgraphs_means" and common attributes:

  • attr(., "dataset"): the original dataset

  • attr(., "variable_label"), attr(., "variable_name")

  • For grouped outputs: attr(., "group_names") and attr(., "group_labels")

  • For multi-variable: attr(., "item_names"), attr(., "item_labels"), attr(., "x_expr")

Details

  • Survey objects are subset to remove NAs in x and grouping variables before grouping, preserving design alignment (ids/strata/weights/replicates).

  • Means and standard errors are computed via survey::svymean(). CIs are Wald intervals using either z (df = Inf) or t (finite df) critical values.

  • sd is the weighted population SD and n is the sum of weights. Both are used in the SE and CI calculations.

  • Group output order follows factor levels for factors, alphabetical for characters, and ascending for numerics.

  • Value labels for grouping variables are preserved in the output by factorizing keys against original labels; variable-level labels are copied to output columns.

Methods

  • get_means.survey.design(): Complex survey designs. For more information visit survey::svydesign() and srvyr::as_survey_design().

  • get_means.svyrep.design(): Replicate-weight survey designs. For more information visit survey::as.svrepdesign() and srvyr::as_survey_rep()

  • get_means.tbl_svy()``: Unwraps srvyr::tbl_svy and dispatches to the appropriate survey method. For more info, visit srvyr::as_survey_design()andsrvyr::as_survey_rep()`.

  • get_means.data.frame(): Non-design summary with optional weights.

Examples

# Setup example data
library(dplyr)

# Let's calculate the overall average score for trad_n
get_means(test_data, trad_n)
#> # A tibble: 1 × 5
#>    mean    sd     n conf_low conf_high
#> * <dbl> <dbl> <dbl>    <dbl>     <dbl>
#> 1  4.22  3.87   250     3.74      4.70

# it also works if x is a string
get_means(test_data, "trad_n")
#> # A tibble: 1 × 5
#>    mean    sd     n conf_low conf_high
#> * <dbl> <dbl> <dbl>    <dbl>     <dbl>
#> 1  4.22  3.87   250     3.74      4.70

# Let's do that again but add weights
get_means(test_data, trad_n, wt = wts)
#> # A tibble: 1 × 5
#>    mean    sd     n conf_low conf_high
#> * <dbl> <dbl> <dbl>    <dbl>     <dbl>
#> 1  3.97  3.76  245.     3.50      4.45

# the wt argument can also be in quotes like this
get_means(test_data, "trad_n", wt = "wts")
#> # A tibble: 1 × 5
#>    mean    sd     n conf_low conf_high
#> * <dbl> <dbl> <dbl>    <dbl>     <dbl>
#> 1  3.97  3.76  245.     3.50      4.45

# Now let's do the average score for different education levels
get_means(test_data, trad_n, edu_f, wts)
#> # A tibble: 4 × 6
#>   edu_f                mean    sd     n conf_low conf_high
#> * <fct>               <dbl> <dbl> <dbl>    <dbl>     <dbl>
#> 1 High School or Less  4.14  3.59  69.8     3.29      5.00
#> 2 Some College         3.93  3.96  86.3     3.08      4.78
#> 3 Bachelor's Degree    4.23  3.83  49.7     3.14      5.32
#> 4 Graduate Degree      3.45  3.41  39.4     2.35      4.55

# it also works with quotes
get_means(test_data, "trad_n", "edu_f", "wts")
#> # A tibble: 4 × 6
#>   edu_f                mean    sd     n conf_low conf_high
#> * <fct>               <dbl> <dbl> <dbl>    <dbl>     <dbl>
#> 1 High School or Less  4.14  3.59  69.8     3.29      5.00
#> 2 Some College         3.93  3.96  86.3     3.08      4.78
#> 3 Bachelor's Degree    4.23  3.83  49.7     3.14      5.32
#> 4 Graduate Degree      3.45  3.41  39.4     2.35      4.55

# you can also pipe in the `data` argument if you want to do some data
# transformations before you calculate the means. For example, say you want
# to compare the means of `trad_n` among people who agreed vs disagreed with
# the variable `top`:
test_data %>%
  mutate(top_f2 = make_dicho(top)) %>%
  get_means(trad_n, top_f2, wts)
#> # A tibble: 2 × 6
#>   top_f2    mean    sd     n conf_low conf_high
#> * <fct>    <dbl> <dbl> <dbl>    <dbl>     <dbl>
#> 1 Agree     5.08  3.83  111.     4.36      5.80
#> 2 Disagree  3.05  3.43  134.     2.46      3.64