/ UserTests Public

Compute stats for usability test data

darrellpenta/UserTests

Folders and files

NameName
Last commit message
Last commit date

History

29 Commits

index.rmd

title author output editor_options
Computing Confidence Intervals in R for UX Researchers
Darrell J. Penta, PhD
html_document
keep_md number_sections
true
true
chunk_output_type
inline

Introduction

The functions in this package compute success rates and confidence intervals following Jeff Sauro's excellent write up on MeasuringU, Lewis and Sauro (2006), and this page by Tom Tullis, which includes a downloadable Excel spreadsheet for computing the Adjusted Walk Confidence Interval.

Lewis and Sauro (2006, p. 144) offer several recommendations:

1. Always compute a confidence interval, as it is more informative than a point estimate. For most usability work, we recommend a 95% adjusted-Wald interval (Sauro & Lewis, 2005).

2. If you conduct usability tests in which your task completion rates typically take a wide range of values, uniformly distributed between 0 and 1, then you should use the LaPlace method. The smaller your sample size and the farther your initial estimate of p is from .5, the more you will improve your estimate of p.

3. If you conduct usability tests in which your task completion rates are roughly restricted to the range of .5 to 1.0, then the best estimation method depends on the value of x/n.

• If x/n ≤ .5, use the Wilson method (which you get as part of the process of computing an adjusted-Wald binomial confidence interval).
• If x/n is between .5 and .9, use the MLE. Any attempt to improve on it is as likely to decrease as to increase the estimate’s accuracy.
• If x/n ≥ .9, but less than 1.0, apply either the LaPlace or Jeffreys method. DO NOT use Wilson in this range to estimate p, even if you have computed a 95% adjusted-Wald confidence interval! (3d) If x/n = 1.0, use the Laplace method.
1. Always use an adjustment when sample sizes are small (n<20). (It does no harm to use an adjustment when sample sizes are larger.)

The equation for the Adjusted-Wald confidence interval is given in (1)

(1)

$$\hat{p}{adj} \pm z{\alpha} \times \sqrt{\frac{\hat{p}{adj}(1-\hat{p}{adj})}{n_{adj}}}$$

$n$ = total trials
$\hat{p}$ = proportion of success trials
$z_{\alpha}$ = the desired critical z-value (defaulting to 1.96 in the success_rate() function of this package)
$\hat{p}{adj} = \left(\frac{(n \times \hat{p} + z{\alpha}^2/2)}{n + z_{\alpha}^2} \right)$
$n_{adj} = n + z_{\alpha}^2$

Point estimators

In addition to the Wilson method, which is used in computing the Adjusted-Wald binomial confidence interval, Lewis and Sauro (2006) also describe the use of the Laplace and Maximum Likelihood Estimate (MLE) for point estimation.

Laplace Method

The equation for the Laplace method is given in (2)

(2) $$\frac{(x + 1)}{(n + 2)}$$

$x$ = the observed number of success trials
$n$ = the total number of trials

Maximum Likelihood Estimate (MLE)

The equation for the Maximum Likelihood Estimate method is given in (3)

(3) $$\frac{x}{n}$$

$x$ = the observed number of success trials
$n$ = the total number of trials

Working with the UserTests functions

Install packages

Install and load the UserTests package and some other packages.

#install.packages("tidyverse", repos = "http://cran.us.r-project.org" )
#install.packages("devtools", repos = "http://cran.us.r-project.org")
#devtools::install_github(repo = "darrellpenta/UserTests")

library(tidyverse) #For importing and exporting data, wrangling data, making figures
library(devtools) #For installing the UserTests package from GitHub
library(UserTests)

Quick Analyses

If you just want to run analyses on a single task for which you know the success rate and the number of trials, provide that information to the UserTests::success_rate function, as below.

mydata <-
success_rate(.success=7, .trials=17)
mydata
##   successes trials orig.succ.pct estimator success.pct low.ci.pct
## 1         7     17         41.18    Wilson        42.8      21.56
##   hi.ci.pct
## 1     64.05


The results returned indicate:

1. successes
The total number of success
2. trials
The total number of trials
3. orig.succ.pct
The raw success rate as a percentage
4. estimator
The name of the method used to adjust the success rate (see Lewis & Sauro, 2006).
5. success.pct
The adjusted success rate as a percentage
6. low.ci.pct
The lower confidence limit as a percentage
7. hi.ci.pct
The upper confidence limit as a percentage

Analyzing larger data sets

Start by properly preparing your data file, which should be saved in.csv format. The data file should have either 3 or 4 columns, depending upon whether one or two test groups are being analyzed.

In both cases, three of the columns should be: Participant (numeric value), Task (numeric value), and Success (numeric value, coded as 1=success, 0=failure). The fourth (optional) column should be Group (character/text value). See table 1:

Table 1. Example task completion data set

1 1 1 US
1 1 0 THEM
1 2 1 US
1 2 1 THEM
2 1 1 US
2 1 0 THEM
2 2 0 US
2 2 1 THEM
... ... ... ...

Import the data

Next, import the .csv file into R using the readr package.

mydata<-
readr::read_csv("sample data/sample_data_2.csv")
## Parsed with column specification:
## cols(
##   Group = col_character(),
##   Participant = col_integer(),
##   Success = col_integer()
## )

head(mydata)
## # A tibble: 6 x 4
##   <chr>       <int> <int>   <int>
## 1 US              1     1       0
## 2 US              1     2       1
## 3 US              1     3       0
## 4 US              1     4       1
## 5 US              2     1       1
## 6 US              2     2       1


Create a table of the adjusted completion rate means and confidence intervals

Run the success_rate function on your data set to view a table of summarized data. The head function in the code below is a convenient way to view the first few rows of data.

mytable<-
success_rate(mydata)
head(mytable)
## # A tibble: 6 x 9
## # Groups:   Task, Group [6]
##    Task Group successes trials orig.succ.pct estimator success.pct
##   <int> <chr>     <int>  <int>         <dbl> <chr>           <dbl>
## 1     1 THEM          4      8          50   MLE              50
## 2     1 US            4      8          50   MLE              50
## 3     2 THEM          5      8          62.5 MLE              62.5
## 4     2 US            7      8          87.5 MLE              87.5
## 5     3 THEM          2      8          25   Wilson           33.1
## 6     3 US            5      8          62.5 MLE              62.5
## # ... with 2 more variables: low.ci.pct <dbl>, hi.ci.pct <dbl>


Export the table

You can export the table as a .csv file. Just include the path to the location where you want to save the file to the path argument in the write_csv function, as in the example below.

readr::write_csv(mytable, path = "..\MyDesktop\MyUsabilityStudy\completion-rates.csv")

Create a figure

If you need a figure, use the comp_figure function. You can overwrite the default labels by providing your own to the appropriate arguments in the function.

myfigure <-
success_rate_fig(mytable, xlabel="Test Task", ylabel = "Success (%)", legend_lab="Groups")
myfigure

Saving the figure

To save the figure, provide the output format and path. The easiest way is to combine both of these in one string. Acceptable file formats include:

• eps
• ps
• tex
• pdf
• jpeg
• tiff
• png
• bmp
• svg
• wmf
ggsave("../MyDesktop/usability-test-figure.png")

Visit my website, darrelljpenta.com or email me

Compute stats for usability test data

Releases

No releases published

Packages 0

No packages published