ICA 2017 Presentation: Science Communication Audience Segmentation Analysis for the USA
1. Audiences for Science Communication:
The United States of America
INSERT CAPTAIN AMERICA
AND TONY STARK LEGO
FIGURES HERE
John C. Besley
Ellis N. Brandt Chair
2. Context
• Request from European
colleagues to look at U.S.
data the biennial National
Science Board report on
public opinion about science
(which I write for the NSB)
• Opportunity to explore Latent
Class Analysis/Mixture Modelling
(in Mplus) for sub-group
identification
• Potential value of developing messages
designed to appeal to sub-groups
3. Variables
drawn from
GSS/NSF
Science and
Engineering
Indicators
2016 data
• High quality,
face-to-face
survey
But …
• Limited
questions and
focused
general views
about S&T
(replicated
with 2014 data)
Descriptive Statistics for 2016 Data Used in Model Mean SD
Education (Less than high school [1] to graduate degree [4]) 2.29 .81
Republican (‘Strong Democrat’ [1] to ‘Strong Republican’ [7]) 3.75 1.94
Ideology (‘Extremely liberal’ [1] to ‘Extremely Conservative’ [7]) 4.06 1.49
“Because of [S&T], there will be more opportunities for
the next generation” (‘Strongly disagree’ [1] to ‘Strongly agree’])
3.27 .64
“Science makes our way of life change too fast.”
(‘Strongly agree’ [1] to ‘Strongly disagree’])
2.44 .80
”Even if it brings no immediate benefits, scientific research that advances the
frontiers of knowledge is necessary and should be supported by the federal
government.” (‘Strongly disagree’ [1] to ‘Strongly agree’])
3.14 .68
Supporting scientific research: … [Do] you think we're spending too much money [1]
on it, too little money [3], or about the right amount [2].
2.29 .66
Scientific community: As far as the people running these institutions are concerned,
would you say you have a great deal of confidence [3], only some confidence [2], or
hardly any confidence at all in them [1]?
2.36 .60
Issues about new scientific discoveries. Are you very interested [3],
moderately interested [2], or not at all interested [1]?
2.29 .70
Issues about technologies: Are you very interested [3],
moderately interested [2], or not at all interested [1]?
2.29 .67
Factual science knowledge (9 questions) 5.80 2.02
Categorical Variables
Use Internet as primary source of S&T news 53%
Male (Dichotomous) 42%
N = ~1,299
4. How many latent classes?
38000.00
39000.00
40000.00
41000.00
42000.00
2
Classes
3
Classes
4
Classes
5
Classes
6
Classes
7
Classes
Sample-Size Adjusted BIC)
2014 2016
0.70
0.75
0.80
0.85
2
Classes
3
Classes
4
Classes
5
Classes
6
Classes
7
Classes
Relative Entropy
2014 2016
0.00
0.05
0.10
0.15
0.20
0.25
2
Classes
3
Classes
4
Classes
5
Classes
6
Classes
7
Classes
LMR Adjusted LRT Significance Test
2014 2016
• BIC suggests more classes is better
(lower = better)
• Entropy also suggests more is better
(higher = better, > .80 best)
• LMR Adjusted LRT suggests
A.) 4 is better than 3 classes
B.) 5 is not better than 4 classes,
C.) 6 is better than 5 or 7classes.
(p < .05 means additional class
does not add explanatory meaning)
5. Let’s look at 4, 5, and 6 class
models using only 2016 data …
6. 33%
Low education centrist who
sees little value in S&T; has low
interest and medium knowledge
18%
Low education
centrist Democrat, sees
has some concerns about
S&T but high interest
and medium knowledge
23%
25%
High education, liberal
Democrat Science lover
A Four Class Model …
Relative entropy: .74
VLMR-LRT: p > .00
High education
conservative Republican , sees
value in S&T and has high
interest and knowledge,
medium support and confidence
7. 11%
Low education centrist,
concerned about the
effect of S&T, limited
interest and knowledge
Low education centrist with
some concern about the
value of the science, limited
interest and medium
knowledge
22%
Medium/high education
conservative Republican
with high S&T interest
and knowledge but
medium support for
funding
25% 24%
Low education, centrist
with some concerns but
general support, high
interest and medium
knowledge
17%
High education, liberal
Democrat Science lover
These are the people
science advocates ‘raise
money’ from and make
sure we don’t alienateThese are the people
we need to figure out
how to reach
(or keep on the
sidelines)
A Five Class Model …
Relative entropy: .75
VLMR-LRT: p = .07
8. 16%
Low education centrist
Democrat, sees
little value in S&T;
has low interest
and knowledge
11%
Low education
centrist, sees
little value in
S&T; has limited
interest
and knowledge
Medium education,
conservative
Republican, sees
some value in S&T;
has limited interest
and knowledge
15%
Low education
centrist, sees
little value in S&T;
has limited interest
and knowledge
19%
22%
High education semi-
conservative Republican
S&T lover
15%
High education, liberal
Democrat Science lover
These are the people
science advocates ‘raise
money’ from and make
sure we don’t alienate
These are the people
we need to figure out
how to reach
(or keep on the
sidelines)
A Six Class Model …
Relative entropy: .80
VLMR-LRT: p = .01
9. Conclusions?
As you segment more …
1. Non-supportive group becomes less evident
2. A group of pro-science conservatives emerges
Next steps …
1. Work on analysis with LCA expert
2. Find data for specific issues
(vs. general views about science)
3. Assess whether different messages resonate
with different groups in predictable ways