Experiments
• Uniquely suited to identify cause-effect
relationships
• To study effect of one variable (treatment) on
another (outcome/dependent variable)
• Use a control group to rule out other causes
• Program is the “treatment” in a program
evaluation; desired outcomes are “effect”
• Measure change with vs without the
program, not just before vs after
Uses of Experiments in PRTR
• Effects of information or promotion programs
– on knowledge, attitudes, or behavior.
• Consumer response to marketing mix changes
– price, product, promotion, place
• Effectiveness of various TR interventions
• Impacts of tourism on community/region
– community attitudes, social, economic, and environmental impacts.
• Benefits/Effects of recreation and tourism activity
– physical health, mental health, family bonding, economic impacts, learning, etc..
• Studying preferences
– for landscapes and more generally to measure the relative importance of
different product attributes in consumer choices. e.g. conjoint analysis
Characteristics of a true Experiment
1. Sample equivalent experimental and
control groups
2. Isolate and control the treatment
3. Measure the effect
Pre-test/Post-test with Control
R MB1 X MA1 Experimental group
R MB2 MA2 Control group
R denotes random assignment to groups
X denotes the treatment
Measure of effect = Expmt gp - Control gp
= (MA1-MB1) - (MA2-MB2)
= with vs without
Example
Pre Post
Expmt 75% 90%
Control 70% 80%
Effect = (90-75) - (80-70)
= 15% - 10% = 5%
With vs without the treatment = 5%
Before vs After = 15%
Threats to Internal validity
• * Pre-measurement (Testing) : effect of pre-
measurement on dependent variable (post-test)
• * Selection: nonequivalent experimental & control groups,
(statistical regression a special case)
• * History: impact of any other events between pre- and
post measures on dependent variable
• * Interaction: alteration of the “effect” due to interaction
between treatment & pre-test.
• Maturation: aging of subjects or measurement procedures
• Instrumentation: changes in instruments between pre and
post.
• Mortality: loss of some subjects
Threats to external validity
• Reactive error - Hawthorne effect - artificiality of
experimental situation
• Measurement timing - measure dependent variable at
wrong time, miss effect.
• Surrogate situation: using population, treatment or
situation different from “real” one.
Quasi-experimental designs
• Ex post facto (after the fact)
• No control group
• Subjects self-select to be in expmt group
1. Travel Bureau compares travel inquiries in 1991 and 1994
to evaluate 1992 promotion efforts.
2. To assess effectiveness of an interpretive exhibit, visitors
leaving park are asked if they saw exhibit or not, Two
groups are compared relative to knowledge, attitudes etc.
Lab vs Field Experiments
Internal vs External Validity
• Internal validity - are findings correct for the
particular subjects & setting
• External validity - can we generalize results
to other similar situations/populations?
• Lab Expmt: high internal validity, low external
• Field Expmt: high external validity, low
internal
Ad Evaluation -Woodside Example
• Design: 30,000 magazine subscribers, randomly assign
10K to each of three groups A, B and C.
• Treatments: 2 expmt’l groups, 1 control
– A- fun in sun message
– B – relax with family message
– C – no ad , control group
• Measures of effect:
– Total inquiries received
– Unaided ad recall via phone survey of 3,000 subscribers
– Expenditures of predicted visitors from each group (phone survey)
Results
Measure of effect A B C
Inquiries/1000 subscribers 30 10 5
Unaided awareness of destin. 12% 4% 2%
Party visits/1000 subscribers 9.0 2.0 .5
Spending per trip $400 $400 $200
Total spending/1,000 subsc. $3,600 $ 800 $ 100
Net tax revenue (10%) per K $360 $80 $10
Ad costs / 1,000 subscribers $40 $40 $0
Net tax/ ad $ (ROI) $9.0 $2.0 -
Tax revenue/ad cost $320 $40 $10
Recommendations
• A-B-C Copy Split
• Large sample sizes – 1,000 plus
• Compare alternatives with each other and to
no ad - A to B and A/B to C
• Track multiple measures of impact/effect
• Gather spending to estimate ROI
Pricing Expmt- Bamford/Manning
• Design: Vary campsite pricing for prime campsites at
Vermont State
• Treatments: Price differentials of $1-$5
– Assign state parks to treatment groups
• Measures of effect:
– Percent choosing prime sites
– Campsite occupancy shift index (compare with previous year)
– Revenue generated
– Equity – acceptance of policy, income group differences
Pricing Expmt Results
• Occupancy shift of 5% for each $1 differential
• Pct choosing prime = 54 - .5* Pctage Price Increase
– E.g. $ 0 differential – 54% choose prime
– 10% differential - 49% ; 20% diff - 44%
• Revenue increase of 4 -22%
• Small differences in income groups
– Pct choose prime 20% for L, 25% M 26% H
– Fee Fair? 49% L, 51% M , 60% H