(based on Shneiderman Chapter 10)

System response time: elapsed time from initiation of activity (e.g. Enter key or OK button) until system response begins to appear.

Display rate: rate at which system response is displayed to user (e.g. in CPS for text)
(for most examples, we assume delay is negligible -- becomes a major factor for web interaction!)

User think time: elapsed time from completion of system response until next initiation of activity
(time for planning and specifying next activity)

Simple productivity model: alternating and non-overlapping periods of think time and response time.

Realistic productivity model: user plans ahead so that think time and response time overlap. This can be further refined by including display rate, and partitioning think time into planning and specifying next activity.

What factors affect display rate in networked environment?
(modem speed, network speed, network congestion)

Compare ASCII versus bitmapped
(given textual information, which has higher effective display rate given equal actual display rate?)

What are implications of this for web page design? For browser design?
(Try to place all textual material in ASCII form, load text before images)

Task: Manipulate some files from your UNIX account using CUI (command). Each command depends on response to previous command. About 20 commands are required (assuming no typing or mental errors).

Scenario 1: response time for each command is average 1 second.

Scenario 2: response time for each command is average 10 seconds.

Question: what is the relationship between Response Time and Productivity (e.g. length of time to complete the task)? Assume:

• 5 seconds average to type a command.
• 2 seconds average think time per command before starting to type the next one.

Calculate task completion time based on non-overlapped think time (think then type then wait).

Calculate task completion time based on overlapped think time (can plan while awaiting response)

Task: Compile a program

Scenario 1: Fall 1995, SMSU, CSC 131, C++.

• How many programming assignments did you have in CSC 131?
  (Those of you who took 131 with me had 10.)
• We were using either Visual C++ 2.0 or Turbo C++ 3.0. Guess the average response time for a compile command.
  (Probably no more than 10 seconds).

Scenario 2: Fall 1972, BGSU, CS 101, Fortran.

• How many programming assignments did I have?
  (6, in a 10-week quarter. Prorate to 16-week semester: 16/10 = x/6, x=9.6)
• We were using Fortran 4 compiler on campus mainframe IBM 360/75. Input was on punched cards, output on printout. Guess the average response time for a compilation.
  (1 to 8 hours, average 4)

Question: what is the relationship between Response Time and Productivity?

Probable results:

In example 1, productivity was directly related to response time (longer means less productive).

In example 2, response time from compiler went from 4 hours (14,400 seconds) down to 10 seconds (1440 times faster), yet productivity was about the same (for comparable program, C++ code is about same length as Fortran code).

What differentiates Example 1 from Example 2?

• In example 1, there were no alternative methods for completing the task -- couldn't even batch the commands. In short, no way to adapt the problem solving process!
• In example 2, there are many approaches, plus you can check the code and even trace it manually before submitting to compile. You can adapt the problem solving process to fit the expected response time. Think time may increase dramatically if response time slow.

Some experiments cited by Shneiderman show steady productivity across a range of response times, due to user adjustment of approach. If task has several possible solutions, then user faced with slow response time will have opportunity to develop a shorter solution (less commands) to compensate. One study showed no significant difference in task completion times for fixed response times of 1, 4, 16, and 64 seconds!

"The nature of the task has a strong influence on whether changes in response time alter user productivity." (page 361).

Examples:

You are driving down a street and a child runs out in front of you. What is acceptable delay from the time you foot hits the brake pedal until the brakes take hold?

You click on link to www.microsoft.com to browse for new info. What is acceptable delay from the click until the home page displays?

At Microsoft website, you elect to download IE 5.0 beta. What is acceptable delay until download completed? Suppose you were connected via modem? Suppose you had LAN connection to T1 service?

You are a Unix novice and working through a tutorial problem given step by step in a book. What is an acceptable delay for return of command prompt after each command?

You are a Unix expert and need to delete some files and directories. What is an acceptable delay for return of command prompt after each command?

You are a Unix expert and need to search for some keywords in a set of text files. What is an acceptable delay for system response?
 

Acceptability of response time varies with:

• Task
• Past experience
• Expectations

Expectation based on

• User task knowledge
• Task complexity
• Past experience
• Individual tolerance for delay

Individual tolerances are huge, but tolerance generally reduced as familiarity increases.

Is there such a thing as a too-fast response?
(can cause anxiety in experienced user -- "what happened?")

Response Time Variability

A number of studies are cited.

• 1968 study by Robert Miller
• Response time: range of 2 to 4 seconds
• Results: 75 of subjects could detect 8% time variation

• 1981 study by Tom Goodman and Robert Spence
• task: manipulate displayed graph using light pen
• response time: mean 1 second, min .2 seconds, max 1.8 seconds
• variations: 3 levels based on "normal" distr: std dev of .2, .4, .8 seconds.
• Results: No significant changes in performance as variation increased

• 1987 study by Werner Kuhmann et al.
• Task: "detection and correction"
• Response times: experiment with short vs. long and fixed vs. variable
• Fixed 2 seconds vs. variable with mean 2 sec, min .5, max 5.75.
• Fixed 8 seconds vs. variable with mean 8 sec, min 2, max 22.81.
• Result: 2 second mean caused higher error rate than 8 second mean.
• Result: No difference in fixed vs. variable, at either level.

Shneiderman Summary (p 366)

• "modest variations in response time (plus or minus 50 percent of the mean) appear to be tolerable and to have little effect on performance."
• "As the variability grows, performance speed may decrease some."
• "Frustration emerges only if delays are unusually long -- at least twice the anticipated time."
• "anxiety about an erroneous command may emerge only if the response time is unusually short -- say, less then one-quarter of the anticipated time."

• 7+-2 comes from George Miller 1956 paper in Psychological Science.
• Refers to number of chunks, where chunk size depends on familiarity with material.
• Novices work better with smaller chunks.
• Chunks can be held in short term memory for 15-30 seconds.
• Distractions and anxiety reduce the capacity of short term and working memory.

Error rates increase:

• If task complexity requires transfer of intermediate info to long term memory
• If result must be held for a long time before being applied
• As system response time increases (beyond a certain threshold). Depends on task and user.

Error rates decrease:

• If user has low anxiety
• If user knows task OA
• If no delay between solution and application
• If no distractions

Other psychological factors resulting from response time:

• novices prefer slower response and may be more productive (do not feel rushed)
• rapid response preferred if error penalty is small
• rapid response preferred if past interactions had rapid response

Last reviewed: 17 November 1998