Who Cares About the Arts?

1
Who Cares About the Arts?

• Predicting Formal Arts Participation from Survey
  Data

Angela Han ECE 539 December 2005
2
Project Objective

• Apply pattern classifier neural network to arts
  marketing survey data
• Use neural network as a predictive model to
  identify potential arts patrons

3
Data

• Americans and the Arts (1992)
• Telephone survey administered February 1992
• 1500 adults across the United States
• Asked 160 questions about participation,
  opinions, socialization, and support of the arts
• Data stored on The Cultural Policy and the Arts
  National Data Archive www.cpanda.org

4
Inputs

• 6 categories of survey questions
• Demographics
• Opinions on art, leisure, and artists
• Participation in arts and leisure activities
• Barriers to participation
• Arts on TV
• Arts socialization
• Support for arts and culture
• Not all questions were used as inputs only
  measurable ones (33 out of 160 questions)

5
Inputs

• Sample questions used for inputs
• What is the last grade or level of school you
  completed?
• Approximately how often did you go to the movies
  in the past 12 months?
• How often would you estimate you buy compact
  discs, tapes, records, or recordings of classical
  music do you buy classical music or recordings
  frequently, every once in a while, only
  occasionally, or almost never?
• Over the past 12 months, have you personally or
  your immediate family contributed any money to an
  arts organization or an arts fund, or not?

6
Outputs

• 6 survey questions were modified to become yes/no
  outputs for neural network
• Not counting any performances given by your
  children in
• connection with school or classes, approximately
  how
• many times in the past 12 months did you go to
• live theater performances
• live classical music performances
• live performances of opera or musical theater
• live performances of dance (ballet, modern,
  folk/ethnic, or jazz)
• art museums or art galleries
• science, or natural history museums or a history
  museum

7
The Model

• A back propagation multilayer perceptron model
  was developed using the bp.m Matlab program from
  the course
• The following parameters were used
• L2, hidden layer10 neurons
• alpha0.01, mu0.8
• epoch size100, max epochs2000.

8
The Results

• bp.m was run for each of the six questions data
  set was the same except for the outputs. Crate
  and error were calculated.

Theatre Music Opera Dance Art Science
crate 54.33 84.33 87.00 86.87 58.80 74.73
Error 18.25 10.16 7.61 8.73 17.04 74.73

• These are not the most ideal results!

9
Discussion

• The data is possibly flawed
• Example 1 respondents were asked if they
  purchased classical music recordings
  frequently, every once in a while, only
  occasionally, almost never, and never.
• All of these choices are subjective, similar
  purchasing habits could be placed in different
  categories.
• Example 2 respondent could respond positively to
  attending classical music concerts in the past 12
  months when he/she had actually been attending
  jazz concerts, or could respond to attending an
  art museum when it was really a natural history
  museum.

10
Discussion

• Data is not properly adjusted
• mean, variance, correlation not adjusted for
• further linear transformation of feature vectors
  may be necessary
• further transformations my be necessary to adjust
  for categorical nature
• SVD may eliminate more features

11
Next Steps

• Further analysis of data
• Further adjustments to MLP structure
• Examine other pattern classifiers
• K-nearest neighbor most intuitive
• Compare with marketing research regression models