Presentation on concept creep research findings: "The semantic inflation of 'trauma' in psychology". Associated publication

1. The semantic inflation of
‘trauma’ in psychology
Naomi Baes, Ekaterina Vylomova, Michael J.
Zyphur, and Nick Haslam
The University of Melbourne, School of
Psychological Sciences
1

2. Trauma creep in everyday language
2

3. Concepts related to harm (abuse, addiction, bullying, prejudice, trauma)
have expanded their meanings to include:
• new kinds of phenomena (horizontal creep)
• less severe phenomena (vertical creep)
Concept creep
Haslam (2016) 3

4. Consequences of concept creep
Negative Implications
• Increases political polarization
• Limits free speech
• Perceive harmful event with
less gravity (Dakin et al., 2022)
Positive Implications
• Promotes social change
• Labels for dubious behaviour
• Extends treatment to people
who were previously denied it
Haslam et al. (2020) review 4

5. Causes of concept creep
Note. Adapted from Haslam et al. (2020) review 5
Horizontal
Individual differences
(McGrath et al., 2019)
• Empathic
• Politically liberal
• Endorse harm-based morality
CONCEPT CREEP
Vertical
Motivated harm inflation (society)
Declining harm prevalence (society)
Rising harm sensitivity (culture)

6. • Reduced severity DSM-III (1980)
• Indirect experiences DSM-III-R (1987)
• Subjective component DSM-IV (1994)
The concept creep of trauma
Traumatic event = “rape”, “assault”, “military combat”, “natural disasters”
Traumatic event = more items, like “business loss”, “marital conflict”
6*

7. “sexual”
“physical” ‘trauma’
Rationale
Year Horizontal creep (T)
(Vylomova et al., 2019; Vylomova & Haslam, 2021)
7
1980s—2010s
“psychological”
“stress”
Trauma has come to be used in a greater range of semantic contexts:

8. Rationale
Trauma frequency
Horizontal creep (T)
(Haslam et al., 2021)
8
Year Horizontal creep (T)
(Vylomova et al., 2019; Vylomova & Haslam, 2021)

9. Trauma frequency
Horizontal creep (T)
(Haslam et al., 2021)
9
Year Horizontal creep (T)
(Vylomova et al., 2019; Vylomova & Haslam, 2021)
Rationale: Evaluating vertical creep
“physical” “psychological”
“stress”
“sexual”

10. Rationale: Evaluating vertical creep
Has ‘trauma’ come to be used in less severe contexts?
10
Trauma frequency
Horizontal creep (T)
(Haslam et al., 2021)
Year Horizontal creep (T)
(Vylomova et al., 2019; Vylomova & Haslam, 2021)
Vertical creep?
Vertical creep?

11. The present study
Aims: Develop new method to evaluate vertical expansion of harm-related
concepts & apply it in a corpus study of trauma using psychology abstracts
Hypothesis 1: The severity of the meanings of (a) words and (b) nouns co-
occurring with trauma has declined in the study period
Hypothesis 2: These declines will be causally influenced by the rising
relative frequency of the trauma concept
11

12. Method: Materials and procedure
12
1. Psychology corpus: 133,082,240 tokens in 871,340 abstracts (875 journals) published 1970-2017
MATERIAL

13. Operationalizing the meaning of ‘trauma’
[ w5 w4 w3 w2 w1 ‘trauma’ w1 w2 w3 w4 w5 ]
13

14. Method: Materials and procedure
14
1. Psychology corpus: 133,082,240 tokens in 871,340 abstracts (875 journals) published 1970-2017
2. Trauma Collocates: 40,708 English words
appearing in ±5-word context window of trauma
MATERIAL
3. Warriner Norms (2013): Valence and Arousal
ratings for 13,915 English lemmas

15. Severity = Negative Valence + High Arousal
* = reverse-scored 15
Severity index = 2 (low severity) —18 (high severity)
Valence scale* = 1 (happy) — 9 (unhappy)
happy completely unhappy
pleased annoyed
unsatisfied satisfied
contented melancholic
Arousal scale = 1 (calm) — 9 (excited)
relaxed stimulated
calm excited
sluggish frenzied
dull jittery

16. Method: Materials and procedure
16
1. Psychology corpus: 133,082,240 tokens in 871,340 abstracts (875 journals) published 1970-2017
3. Warriner Norms (2013): Valence and Arousal
ratings for 13,915 English lemmas
2. Trauma Collocates: 40,708 English words
appearing in ±5-word context window of trauma
MATERIAL
WORDS
Word List: 3,918 unique sentence-level trauma collocates, with associated severity ratings

17. 17
Method: Materials and procedure
1. Psychology corpus: 133,082,240 tokens in 871,340 abstracts (875 journals) published 1970-2017
3. Warriner Norms (2013): Valence and Arousal
ratings for 13,915 English lemmas
2. Trauma Collocates: 40,708 English words
appearing in ±5-word context window of trauma
WORDS
MATERIAL
NOUNS
Word List: 3,918 unique sentence-level trauma collocates, with associated severity ratings (1974-2017)

18. 18
Method: Materials and procedure
1. Psychology corpus: 133,082,240 tokens in 871,340 abstracts (875 journals) published 1970-2017
3. Warriner Norms (2013): Valence and Arousal
ratings for 13,915 English lemmas
2. Trauma Collocates: 40,708 English words
appearing in ±5-word context window of trauma
WORDS
3,396 unique sentence-level trauma collocates appearing as nouns at least once
Noun List: 2,117 unique sentence-level trauma collocates occurring mainly as nouns, with severity ratings
MATERIAL
NOUNS
Word List: 3,918 unique sentence-level trauma collocates, with associated severity ratings (1974-2017)

19. Method: Index construction
Trauma concept
frequency
Sum(relative frequency of
“trauma”, “traumatiz(+s)e,
“traumatic”)
Year(1974-2017)
19
Trauma severity (words)
sum(severity ratings * word
repetition) by year/
sum(repetitions) by year
Trauma severity
(nouns)

20. Method: Analytic Strategy
Trauma concept
frequency
H1a
H1b
H2a
H2b
20
Trauma severity
(nouns)
Year(1974-2017)
H1: Severity of the meanings of (a) words & (b) nouns co-occurring with trauma has declined in study period
H2: These declines will be causally influenced by the rising relative frequency of the trauma concept
Trauma severity
(words)

21. H1: Significant year effects from 1974-2017
β = -.37; p = .013
β = -.30; p = .048
The severity of trauma-related words (H1a) and nouns (H1b) declined as a function of year:
21

22. Trauma has risen in prominence in the psych corpus
22

23. When the relative frequency of trauma was high, its
severity tended to be low
Note. Each dot represents a year in the time series data 23
r(42) = -.32; p = .032
r(42) = -.40; p = .008

24. H2a: Predictive effect for rising trauma relative frequency on the
declining severity index of trauma-related words
Overall model stats = F(9, 25) = 5.49, p < .001, Adj. R2 = 0.54 24
Best fitted model produced lag weights for Frequency (2 lags) and Severity (6 lags)
† = total predictive effect estimated by summing all short-run coefficients for predictor/outcome pair

25. H2b: Predictive effect for rising trauma relative frequency on the
declining severity index of trauma-related nouns
Overall model stats = F(11, 23) = 4.10, p = .002, Adj. R2 = .50 25
Best fitted model produced lag weights for Frequency (4 lags) and Severity (6 lags)
† = total predictive effect estimated by summing all short-run coefficients for predictor/outcome pair

26. The present study
Aims achieved: Developed a new method to
assess the vertical form of concept creep &
tested it in a corpus study of trauma using
psychology article abstracts
Hypotheses supported (1974—2017):
• H1: Year  Trauma severity (w,n)
• H2: Trauma frequency  Trauma severity (w,n)
26*

27. What does this all mean?
H1 findings:
• Consistent with the general claim of
concept creep theory – that trauma is
increasingly used in less severe semantic
contexts (trend in academic discourse)
• Complement previous findings showing
trauma’s horizontal expansion using the
semantic breadth measure (Vylomova et al., 2019;
Vylomova & Haslam, 2021)
H2 findings:
• Demonstrate that frequency is likely a
causal linguistic factor in concept creep
(Haslam et al., 2021)
27*

28. Implications of the present findings for:
Academic psychology
• Vertical expansion of
trauma in psychology
abstracts signals a relaxed
threshold for what
constitutes ‘trauma’ PTSD
over-diagnosis (?), post-traumatic reactions
• Concepts receiving more
attention  more likely to
be applied across several
contexts in the academic
literature (concept creep)
e.g., debate in trauma studies literature
28*
Concept creep theory
• Our method to index the semantic
severity of trauma pioneered a
method to assess vertical creep
• Count-based linguistic model linked to affective ratings
• Demonstrates that concept creep
indeed has two distinct components
• Frequency effect indicates that there
may be a role for linguistic factors in
the concept creep framework (currently
specifies only sociocultural causes)

29. Limitations and future directions
• Link concrete (experience-based)/abstract lemma
dataset (Brysbaert et al., 2014)
• Test year and language effects in general domain
corpus (CoHA+CoCA; Vylomova & Haslam, 2021)
• Explore vertical and horizontal expansion in other
harm-related concepts (e.g., mental disorder, abuse)
• Introduce socio-cultural predictors in ARDL model
(e.g., harm sensitivity, economic hardship)
• Noun list not sufficiently narrow (?)
• Only tested for effects in
psychology corpus
• Possible socio-cultural confounds
29*

30. Our contribution to the literature
In summary
• First study to develop a method to evaluate
vertical semantic expansion in the context of
trauma, laying groundwork for future research
• Established potential causal factor implicated
in concept creep: concept frequency
• Method provides research opportunities to
investigate vertical expansion in other harm-
related concepts (psychology, culture)
30

31. Acknowledgements
Professor Nick Haslam
Dr Ekaterina Vylomova
The Concept Creep Lab Members
31

32. Thank you.
Questions?
Note: Images in the public domain unless [page number]* in which case = Akiane Kramarik. 32