dropR: Dropout Analysis by Condition

Description

Analysis and visualization of dropout between conditions in surveys and (online) experiments. Features include computation of dropout statistics, comparing dropout between conditions (e.g. Chi square), analyzing survival (e.g. Kaplan-Meier estimation), comparing conditions with the most different rates of dropout (Kolmogorov-Smirnov) and visualizing the result of each in designated plotting functions. Sources: Andrea Frick, Marie-Terese Baechtiger & Ulf-Dietrich Reips (2001) https://www.researchgate.net/publication/223956222_Financial_incentives_personal_information_and_drop-out_in_online_studies; Ulf-Dietrich Reips (2002) "Standards for Internet-Based Experimenting" doi:10.1027//1618-3169.49.4.243.

Author(s)

Maintainer: Annika Tave Overlander annika-tave.overlander@uni.kn

Authors:

• Matthias Bannert bannert@kof.ethz.ch

Other contributors:

• Ulf-Dietrich Reips [contributor]

See Also

Useful links:

• https://iscience-kn.github.io/dropR/

• https://github.com/iscience-kn/dropR

Add Dropout Index to a Data.Frame

Description

Find drop out positions in a data.frame that contains multiple questions that had been asked sequentially. This function adds the Dropout Index variable do_idx to the data.frame which is necessary for further analyses of dropout.

Use this function first to prepare your dropout analysis. Then, keep going by creating the dropout statistics using compute_stats().

Usage

add_dropout_idx(df, q_pos)

Arguments

Details

Importantly, this function will start counting missing data at the end of the data frame. Any missing data which is somewhere in between, i.e. a single item that was skipped or forgotten will not be counted as dropout. The function will identify sequences of missing data that go until the end of the data frame and add the number of the last answered question in do_idx.

Therefore, the variables must be in the order that they were asked, otherwise analyses will not be valid.

Value

Returns original data frame with column do_idx added.

Source

R/add_dropout_idx.R

See Also

compute_stats() which is usually the next step for dropout analysis.

Examples

dropout <- add_dropout_idx(dropRdemo, 3:54)

Compute Dropout Statistics

Description

This is the second step in conducting dropout analysis with dropR. Outputs all necessary statistics to analyze and visualize dropout, such as the sample size N of the data (and in each condition if selected), cumulative dropout and remaining participants in absolute numbers and percent. If no experimental condition is added, the stats are only calculated for the whole data in total.

Usage

compute_stats(df, by_cond = "None", no_of_vars)

Arguments

Value

A data frame with 6 columns (q_idx, condition, cs, N, remain, pct_remain) and as many rows as questions in original data (for overall data and if conditions selected again for each condition).

Examples

do_stats <- compute_stats(df = add_dropout_idx(dropRdemo, 3:54),
by_cond = "experimental_condition",
no_of_vars = 52)
  

Compute Chisq-Test Given a Question Position

Description

This function performs a chi-squared contingency table test on dropout for a given question in the data. Note that the input data should be in the format as computed by compute_stats(). The test can be performed on either all conditions (excluding total) or on select conditions.

Usage

do_chisq(do_stats, chisq_question, sel_cond_chisq, p_sim = TRUE)

Arguments

Value

Returns test results from chisq.test between experimental conditions at defined question.

See Also

add_dropout_idx() and compute_stats() which are necessary for the proper data structure.

Examples

do_stats <- compute_stats(add_dropout_idx(dropRdemo, 3:54),
by_cond = "experimental_condition",
no_of_vars = 52)

do_chisq(do_stats, 47, c(12, 22), TRUE)

Kaplan-Meier Survival Estimation

Description

This function needs a data set with a dropout index added by add_dropout_idx(). The do_kpm function performs survival analysis with Kaplan-Meier Estimation and returns a list containing survival steps, the original data frame, and the model fit type. The function can fit the survival model either for the entire data set or separately by a specified condition column.

Usage

do_kpm(df, condition_col = "experimental_condition", model_fit = "total")

Arguments

Value

Returns a list containing steps (survival steps extracted from the fitted models), d (the original data frame), and model_fit (the model fit type).

See Also

survival::Surv() used to fit survival object.

Examples

demo_kpm <- do_kpm(df = add_dropout_idx(dropRdemo, 3:54),
condition_col = "experimental_condition",
model_fit = "total")

head(demo_kpm$steps)

Compute Kolmogorov-Smirnov Test for most extreme conditions

Description

This test is used for survival analysis between the most extreme conditions, so the ones with the most different rates of dropout. This function automatically prepares your data and runs stats::ks.test() on it.

Usage

do_ks(do_stats, question)

Arguments

Value

Returns result of Kolmogorov-Smirnoff test including which conditions have the most different dropout rates.

Examples

do_stats <- compute_stats(df = add_dropout_idx(dropRdemo, 3:54),
by_cond = "experimental_condition",
no_of_vars = 52)

do_ks(do_stats, 52)

Dropout Odds Ratio Table

Description

This function calculates an Odds Ratio table at a given question for selected experimental conditions. It needs data in the format as created by compute_stats() as input.

Usage

do_or_table(do_stats, chisq_question, sel_cond_chisq)

Arguments

Value

Returns a Matrix containing the Odds Ratios of dropout between all selected conditions.

See Also

compute_stats()

Examples

do_stats <- compute_stats(df = add_dropout_idx(dropRdemo, 3:54),
by_cond = "experimental_condition",
no_of_vars = 52)

do_or_table(do_stats, chisq_question = 51, sel_cond_chisq = c("11", "12", "21", "22"))

Calculate Steps for Uneven Data Points

Description

The do_steps function calculates steps for data points represented by numbers of questions from the original experimental or survey data in x and remaining percent of participants in y.

Usage

do_steps(x, y, return_df = TRUE)

Arguments

Details

Due to the nature of dropout/ survival data, step functions are necessary to accurately depict participants remaining. Dropout data includes the time until the event (a.k.a. dropout at a certain question or time), so that changes in remaining participants are discrete rather than continuous. This means that changes in survival probability occur at specific points and are better represented as steps than as a continuum.

Value

Returns a data frame or a list containing the modified x and y values.

Examples

x <- c(1, 2, 3, 4, 5)
y <- c(100, 100, 95, 90, 85)
do_steps(x, y)

# Using the example dataset dropRdemo

do_stats <- compute_stats(df = add_dropout_idx(dropRdemo, 3:54),
by_cond = "experimental_condition",
no_of_vars = 52)

tot_stats <- do_stats[do_stats$condition == "total", ]
do_steps(tot_stats$q_idx, tot_stats$pct_remain)

Demo Dataset for Dropout in an Online Survey

Description

Simulated demo data set for dropout in a survey.

Format

A data frame with 246 rows and 54 variables (in the order they were presented in the fictional survey).

obs_id

Observation ID

experimental_condition

experimental condition

vi_1

item 1

vi_2

item 2

vi_3

item 3

vi_4

item 4

vi_5

item 5

vi_6

item 6

vi_7

item 7

vi_8

item 8

vi_9

item 9

vi_10

item 10

vi_11

item 11

vi_12

item 12

vi_13

item 13

vi_14

item 14

vi_15

item 15

vi_16

item 16

vi_17

item 17

vi_18

item 18

vi_19

item 19

vi_20

item 20

vi_21

item 21

vi_22

item 22

vi_23

item 23

vi_24

item 24

vi_25

item 25

vi_26

item 26

vi_27

item 27

vi_28

item 28

vi_29

item 29

vi_30

item 30

vi_31

item 31

vi_32

item 32

vi_33

item 33

vi_34

item 34

vi_35

item 35

vi_36

item 36

vi_37

item 37

vi_38

item 38

vi_39

item 39

vi_40

item 40

vi_41

item 41

vi_42

item 42

vi_43

item 43

vi_44

item 44

vi_45

item 45

vi_46

item 46

vi_47

item 47

vi_48

item 48

vi_49

item 49

vi_50

item 50

vi_51

item 51

vi_52

item 52

Source

dropRdemo Demo data for dropout.

Compute Odds From Probabilities

Description

Compute odds from probabilities. The function is vectorized and can handle a vector of probabilities, e.g. remaining percent of participants as calculated by compute_stats().

Usage

get_odds(p)

Arguments

Value

Returns numerical vector of the same length as original input reflecting the odds.

Examples

get_odds(0.7)
get_odds(c(0.7, 0.2))

Compute Odds Ratio

Description

Computes odds ratio given two probabilities. In this package, the function can be used to compare the percentages of remaining participants between two conditions at a time.

Usage

get_odds_ratio(a, b)

Arguments

Value

Returns numerical vector of the same length as original input reflecting the Odds Ratio (OR).

See Also

get_odds(), as this is the basis for calculation.

Examples

get_odds_ratio(0.7, 0.6)

Get Steps Data by Condition

Description

The get_steps_by_cond function calculates steps data based on survival model results. This utility function is used inside the do_kpm() function of dropR.

Usage

get_steps_by_cond(sfit, condition = NULL)

Arguments

Value

Returns a data frame containing the steps data, including time, survival estimates, upper confidence bounds, and lower confidence bounds.

See Also

do_kpm()

Test Survival Curve Differences

Description

This function compares survival curves as modeled with do_kpm(). It outputs a contingency table and a Chisq measure of difference.

Usage

get_survdiff(kds, cond, test_type)

Arguments

Value

Returns survival test results as called from survival::survdiff().

Examples

kpm_est <- do_kpm(add_dropout_idx(dropRdemo, 3:54))
get_survdiff(kpm_est$d, "experimental_condition", 0)
get_survdiff(kpm_est$d, "experimental_condition", 1)

Plot Dropout Curves

Description

This functions uses ggplot2to create drop out curves. Please note that you should use add_dropout_idx() and compute_stats() on your data before running this function as it needs a certain data structure and variables to work properly.

Usage

plot_do_curve(
  do_stats,
  linetypes = TRUE,
  stroke_width = 1,
  full_scale = TRUE,
  show_points = FALSE,
  show_confbands = FALSE,
  color_palette = "color_blind"
)

Arguments

Value

Returns a ggplot object containing the dropout curve plot. Using the Shiny App version of dropR, this plot can easily be downloaded in different formats.

See Also

add_dropout_idx() and compute_stats() which are necessary for the proper data structure.

Examples

do_stats <- compute_stats(add_dropout_idx(dropRdemo, 3:54),
by_cond = "experimental_condition",
no_of_vars = 52)

plot_do_curve(do_stats)

Plot a Kaplan Meier Survival Estimation

Description

The plot_do_kpm function generates a Kaplan-Meier survival plot based on the output from the do_kpm() function. It allows for customization of conditions to display, confidence intervals, color palettes, and y-axis scaling.

Usage

plot_do_kpm(
  kds,
  sel_conds = c("11", "12", "21", "22"),
  kpm_ci = TRUE,
  full_scale_kpm = FALSE,
  color_palette_kp = "color_blind"
)

Arguments

Value

Returns a ggplot object containing the Kaplan-Meier survival plot. Using the Shiny App version of dropR, this plot can easily be downloaded in different formats.

Examples

plot_do_kpm(do_kpm(d = add_dropout_idx(dropRdemo, 3:54),
condition_col = "experimental_condition",
model_fit = "total"))

plot_do_kpm(do_kpm(d = add_dropout_idx(dropRdemo, 3:54),
condition_col = "experimental_condition",
model_fit = "conditions"), sel_conds = c("11", "12", "21", "22"))

Plot Most Extreme Conditions to Visualize Kolmogorov-Smirnov Test Results

Description

With this function, you can easily plot the most extreme conditions, a.k.a. those with the most different dropout rates at a certain question. You need to define that question in the function call of do_ks() already, or just call that function directly inside the plot function.

Usage

plot_do_ks(
  do_stats,
  ks,
  linetypes = FALSE,
  show_confbands = FALSE,
  color_palette = c("#E69F00", "#CC79A7")
)

Arguments

Value

Returns a ggplot object containing the survival curve plot of the most extreme dropout conditions. Using the Shiny App version of dropR, this plot can easily be downloaded in different formats.

See Also

compute_stats(), do_ks()

Examples

do_stats <- compute_stats(add_dropout_idx(dropRdemo, 3:54), 
by_cond = "experimental_condition",
no_of_vars = 52)

ks <- do_ks(do_stats, 52)

plot_do_ks(do_stats, ks, color_palette = "gray")

# ... or call the do_ks() function directly inside the plotting function
plot_do_ks(do_stats, do_ks(do_stats, 30))

plot_do_ks(do_stats, ks, linetypes = TRUE, 
show_confbands = TRUE, color_palette = c("red", "violet"))

Start the dropR Shiny App

Description

Starts the interactive web application to use dropR in your web browser. Make sure to use Google Chrome or Firefox for best experience.

Usage

start_app()

Details

The app will give less experienced R users or statisticians a good overview of how to conduct dropout analysis. For more experienced analysts, it can still be very helpful in guiding how to use the package as there are some steps that should be taken in order, which is outlined in the app (as well as function documentation).

Value

No return value; starts the shiny app as a helper to get started with dropout analysis. All app procedures are available as functions.