Data Center Sentiment - Study 3: Public Benefit v External Threat Framing

Data center buildouts face a challenging landscape. As construction expands its footprint across the country, data centers are increasingly acquiring neighbors: everyday Americans in their homes. For millions of Americans, the question of whether a data center belongs in their community has moved from abstract policy debate to a concrete local decision. And increasingly, the answer is no. This resistance follows a predictable pattern that is consistent with construal level frameworks: the benefits of data centers are diffuse and remote, while the costs are immediate and local, the precise conditions under which NIMBY opposition takes hold.

While data center benefits are rendered abstractly relative to their costs, those abstract benefits can take two distinct psychological pathways. Benefits can be framed as promotion-focused — where they constitute growth and advancement — or prevention-focused — where they involve the avoidance of loss, typically in response to a threat. Prevention-focused framing is traditionally considered more persuasive in part because it is better aligned with concrete construals, where costs are salient and urgency is foregrounded. By narrowing psychological distance, prevention framing brings a topic closer to the concrete level where opposition typically lives — and, critically, may be better positioned to compete with it.

Whether framing type can meaningfully shift public support for local data center construction is an open question that produces competing hypotheses. The present study tests this directly by comparing a promotion-framed benefit condition, in which a prospective local data center is described as enabling biomedical research, to a prevention-framed benefit condition, in which a prospective data center is positioned as a necessary response to China’s AI-enabled cyber espionage threats. If prevention framing operates at the same concrete construal level as the costs it must compete with — the land use, the energy burden, the erosion of local quality of life — it may be better positioned to register as a genuine counterweight in the cost-benefit calculus. On this account, support for local data center construction should be higher in the prevention condition than in the promotion condition, because it brings the otherwise abstract benefits down to the same psychological terrain where the costs already live.

If promotion-focused framing outcompetes the prevention-focused framing in increasing tolerance for a local data center build, then it may be the case that opinions about local data centers – and potentially NIMBY reasoning writ large – may not align within the traditional regulatory fit framework (Higgins et al., 2001).

Participants first respond to two moderator items.

Belief in a Dangerous World

(Altemeyer 1988; Duckitt & Fisher, 2003; Perry et al., 2013)

[1 = strongly disagree; 7 = strongly agree]

  1. My knowledge and experience tells me that the social world we live in is basically a safe, stable and secure place in which most people are fundamentally good
  2. It seems that every year there are fewer and fewer truly respectable people, and more and more persons with no morals at all who threaten everyone else
  3. Although it may appear that things are constantly getting more dangerous and chaotic, it really isn’t so. Every era has its problems, and a person’s chances of living a safe, untroubled life are better today than ever before
  4. Any day now chaos and anarchy could erupt around us. All the signs are pointing to it
  5. There are many dangerous people in our society who will attack someone out of pure meanness, for no reason at all
  6. The ‘‘end’’ is not near. People who think that earthquakes, wars, and famines mean God might be about to destroy the world are being foolish
  7. My knowledge and experience tells me that the social world we live in is basically a dangerous and unpredictable place, in which good, decent and moral people’s values and way of life are threatened and disrupted by bad people
  8. Despite what one hears about ‘‘crime in the street,’’ there probably isn’t any more now than there ever has been
  9. If a person takes a few sensible precautions, nothing bad is likely to happen to him or her; we do not live in a dangerous world
  10. Every day as society become more lawless and bestial, a person’s chances of being robbed, assaulted, and even murdered go up and up

Perception of societal threat

[1 = strongly disagree, 7 = strongly agree]

The American way of life is under threat.

Manipulation

Participants were then randomly assigned to one of two conditions.

Biomedical Benefit condition

Imagine a biomedical research institution is interested in building a data center in your community to support AI technology, which will accelerate scientific discovery into life-saving medicines. This research is intended to support Americans’ access to the best and most effective medicines.

Cyber Threat condition

Imagine a cybersecurity firm is interested in building a data center in your community to support AI technology, which will advance cybersecurity defense against Chinese government hackers. In 2024, the Chinese government hacked critical U.S. infrastructure, compromising government wiretapping systems and gaining access to millions of Americans’ sensitive data.

DVs

Participants then responded to two DV measures and a comprehension check

Support for local data center

Would you favor or oppose the construction of this new data center in your local area?

[1 = strongly oppose, 3 = neither/not sure, 5 = strongly favor]

Urgency and Importance

Some things in life are important but not urgent: they matter a great deal but don’t require immediate action. Other things are urgent but not necessarily important: they demand attention now but may not matter as much as highly important things. And some things are both urgent and important: they matter a great deal and require immediate action. The following questions ask you to evaluate AI infrastructure development in the United States on both dimensions.

[1 = strongly disagree, 7 = strongly agree]

  • Expanding data center infrastructure is an important matter for the US.

  • Expanding data center infrastructure is an urgent matter for the US.

Comprehension check

The data center that was proposed in your community was intended to

  • train the next generation of frontier AI models.
  • accelerate research into life-saving medicines.
  • defend against foreign cyberattacks on U.S. infrastructure.
  • expand broadband internet access to rural communities.
show code 
sentiment_study3 = read.csv('dataCenterSentiment_3.csv', stringsAsFactors = TRUE) %>% 
  filter(condition!='') %>% 
  mutate(BDW_1 = BDW_1*-1,
         BDW_3 = BDW_3*-1,
         BDW_6 = BDW_6*-1,
         BDW_8 = BDW_8*-1,
         BDW_9 = BDW_9*-1) %>% 
  mutate(BeliefDangerousWorld = rowMeans(select(.,matches("BDW")), na.rm=FALSE),
         PID = 1:nrow(.),
         compCheck_pass = ifelse(condition == 'BiomedicalBenefit' & compCheck==2, 1, 
                                 ifelse(condition == 'ExternalThreat' & compCheck==3, 1, 0)))

Support, urgency, and importance predicted by condition

show code 
study3_tests = sentiment_study3 %>% select(PID, condition, importance, urgency, support) %>% 
  pivot_longer(3:5, names_to = "variable")

ggerrorplot(data=study3_tests, x='variable',y='value', color='condition')

show code 
nice_table(study3_tests %>% group_by(variable) %>% 
  t_test(value ~ condition, paired = FALSE, var.equal=TRUE) %>% mutate(p = round(p, 3)) %>% 
  inner_join(cohens_d(study3_tests %>% group_by(variable), value~condition, paired=FALSE)) %>%  
  select(variable, statistic, df, p, effsize))
Joining with `by = join_by(variable, .y., group1, group2, n1, n2)`

variable

statistic

df

p

effsize

importance

-1.71

474

.088

-0.16

support

1.12

471

.264

0.10

urgency

-1.00

472

.317

-0.09

Support moderated by BDW

show code 
ggscatter(data=sentiment_study3, x='BeliefDangerousWorld', y='support', color='condition', add = "reg.line", conf.int = TRUE, position= position_jitter(width=0.2))

show code 
nice_table(tidy(lm(data=sentiment_study3, support~condition*BeliefDangerousWorld)))

Term

estimate

std.error

statistic

p

(Intercept)

-0.34

0.09

-3.78

< .001***

conditionExternalThreat

-0.19

0.12

-1.58

.116

BeliefDangerousWorld

-0.10

0.07

-1.39

.164

conditionExternalThreat × BeliefDangerousWorld

0.03

0.10

0.27

.787

Support moderated by perceived threat

show code 
ggscatter(data=sentiment_study3, x='perceived_threat', y='support', color='condition', add = "reg.line", conf.int = TRUE, position= position_jitter(width=0.2))

show code 
nice_table(tidy(lm(data=sentiment_study3, support~condition*perceived_threat)))

Term

estimate

std.error

statistic

p

(Intercept)

-0.33

0.09

-3.51

< .001***

conditionExternalThreat

-0.16

0.14

-1.19

.234

perceived_threat

-0.07

0.05

-1.45

.148

conditionExternalThreat × perceived_threat

0.04

0.07

0.56

.577

Comprehension check

show code 
nice_table(sentiment_study3 %>% group_by(condition, compCheck_pass) %>% tally() %>% mutate(percent = n/sum(n)*100))

condition

compCheck_pass

n

percent

BiomedicalBenefit

0.00

96

40.34

BiomedicalBenefit

1.00

141

59.24

BiomedicalBenefit

1

0.42

ExternalThreat

0.00

56

23.53

ExternalThreat

1.00

182

76.47

Support, urgency, and importance predicted by condition - excluding comp check fails

show code 
study3_tests_compCheckPass = sentiment_study3 %>% filter(compCheck_pass==1) %>% select(PID, condition, importance, urgency, support) %>% 
  pivot_longer(3:5, names_to = "variable")

ggerrorplot(data=study3_tests_compCheckPass, x='variable',y='value', color='condition')

show code 
nice_table(study3_tests_compCheckPass %>% group_by(variable) %>% 
  t_test(value ~ condition, paired = FALSE, var.equal=TRUE) %>% mutate(p = round(p, 3)) %>% 
  inner_join(cohens_d(study3_tests_compCheckPass %>% group_by(variable), value~condition, paired=FALSE)) %>%  
  select(variable, statistic, df, p, effsize))
Joining with `by = join_by(variable, .y., group1, group2, n1, n2)`

variable

statistic

df

p

effsize

importance

-1.07

321

.286

-0.12

support

1.45

319

.147

0.16

urgency

-0.54

320

.591

-0.06