Creating User Interfaces

1
Creating User Interfaces

• Qualitative vs Quantitative research.
• Sampling. Panels.
• Homework Post proposal work on user
  observation study. Next weekReview HTML
  JavaScript

2
Schedule

• Next week HTML5 JavaScript recap
• User observation study presentations week after
  next.
• Be prepared for the first day! Have 1-pager.
• Embedded computers. Plan studies
• Embedded apps presentations.
• Spring break
• VoiceXML 2 weeks
• Teaching project. Various topics.
• Presentations.

3
Research

• Much research in usability is more-or-less
  qualitative
• Observations
• Focus groups
• Monitoring systems MAY apply a metric to
  complaint and act once a threshold is reached.
• Still, there may be reasons for gathering
  quantitative information
• what capacity is required
• Storage
• Simultaneous response
• speeds

4
Panels

• Recruit a panel of people
• Answer questions and/or
• Be willing to be monitored on actions
• Often, open-ended recruiting and/but determine
  critical demographics
• Age
• Gender
• Location
• Device
• ?
• Do need to decide if those who volunteer are
  different from the regular population.

5
Interpret findings

• Assume you have accurate model of the user
  population
• Adjust (normalize) findings

Category Actual pop. Panel Produced Adjusted
Desc. A P X (A/P)X
Young men 10000 103 345 33495
Old men 4000 85 210 9882
Young women 5000 56 87 7768
Old women 4000 61 78 5115
6
Very quick Statistics

• Mean
• Median
• Standard Deviation and Variance
• Normal distribution

7
Sampling

• Done to make an informed estimate of something
  for a large population (of people or things) when
  it is too expensive or difficult to ask every
  person or measure every thing.
• Typical findingWe are 95 that the actual value
  or proportion is within a certain rangex-
  Margin_of_Error lt x lt xMargin_of_Error

8
Example

• Find out how many people think the latest version
  of your program is better than the last.
• Ask N people. Say p is the proportion that said
  yes.
• Margin_of_error
• ztransform square_root((p) (1-p)/N)
• Where ztransform is based on confidence
  level1.96 for 95.

9
Example continued

• N is 1500.
• p is 822/1500 or 54.8
• M 1.96 SQRT((822/1500)(678/1500)/1500)
• M is 2.5
• So we are 95 confident that between 54.8-2.5
  which is (about) 52.3 and54.82.5 which is
  (about) 57.3 think the new system is
  better.

10
Caution

• There is a chance (say 1/20) that the prediction
  is wrong. If you want something less, then choose
  a different confidence level with a different
  z-transform
• Typical choice 99 confidence, multiply by
  2.58Bigger margin means more confident.
• We are

11
Example with different confidence level

• N is 1500.
• p is 822/1500 or 54.8
• M 2.58 SQRT((822/1500)(678/1500)/1500)
• M is 3.3
• So we are 99 confident that between 54.8-3.3
  which is (about) 51.5 and54.83.3 which is
  (about) 58.1 think the new system is better.

12
Warning

• The formula works if the sample is truly random,
  that is
• Every person in the whole population stands the
  same chance as being in the sample.
• Predictions fail when sample isn't random.
• Well-done analysis of election polling works
• Reference Nate Silver

13
Reference

• The Cartoon Guide to Statistics by Larry Gonick
  and Woollcott Smith.
• Consider taking Introduction to Statistics
  course.
• Probably offer Operations Research and Data
  Science next Spring.

14
Panels and/or testing

• When testing for usability, need to evaluate
  costs/benefits of formal testing versus access to
  subjects that will supply more information.
• Comments?

15
Classwork / Homework

• Form teams.
• Plan project.
• Post proposal for user observation study
  (indicating teams)
• Start study
• Next week Review HTML5 and JavaScript
• Processing JS is an option.