Transcript CT ST neutral

Predicting Change-Talk and
Sustain-Talk from Counselor Behavior.
Results from Sequential Analyses of 162
MI-Sessions with Smoking Women Post Partum
Wolfgang Hannöver
Universitymedicine Greifswald
Institute for Medical Psychology
Overview






Background
Aim
Material and Methods
 Data Base
 Coding
 Statistical Analyses
Results
Discussion
Implications for Practice
Sequential Analyses, ICMI 2014
Background

MI effective intervention




Process research




Possibility to „look under the hood“
Testing assumed effects
Support in formulating theoretical model
Technical Hypotheses (Miller & Rose, 2009)




Variability in effect-sizes across studies
Pragmatically developed in practice
Theoretical foundation implicit / post hoc
MI-consistent behavior (MICO) leads to Change Talk (CT)
MI-inconsistent behavior (MIIN) leads to Sustain Talk (ST)
Amount of CT predicts behaviour change
Empirical support for causal chain (Moyers et al. 2009)
Sequential Analyses, ICMI 2014
Background

Meta-Analysis of 12 studies with 1004 subjects
(Magill et al in pr.)

Hypotheses for in-session behaviors

Support for:

more MICO leading to more CT
more MIIN leading to less CT
more MIIN leading to more ST
No support for: more MICO leading to less ST

Hypotheses for in-session behavior to outcomes



Support for:
No support for:
Support for:
ST associated with poor outcomes
CT associated with good outcomes
Positive balance of CT vs. ST associated with good
outcomes
Sequential Analyses, ICMI 2014
Aims

Replication of the hypotheses of in-session behavior using
sequential analysis in a sample of MI-sessions with mothers
post partum who smoked prior to or in pregnancy

Investigate role of single codes
Sequential Analyses, ICMI 2014
Hypotheses











Upon MICO: probability for CT higher than ST or neutral language
Upon MICO: probability for ST lower than CT or neutral language
Upon MIIN: probability for ST higher than CT or neutral language
Upon MIIN: probability for CT lower than ST or neutral language
Upon questions about negative aspects of smoking -> more CT
Upon questions about positive aspects of smoking -> more ST
Upon neutral questions -> more neutral language
Upon
Upon
Upon
Upon
positive reflections
negative reflections
both sided reflections
neutral reflections
->
->
->
->
more
more
more
more
Sequential Analyses, ICMI 2014
CT
ST
CT
neutral language
Data base

MI-based intervention to





Initiate smoking cessation
Prevent return to regular smoking in women post partum
Intervention postponed relapse for six months
No significant increase of initiation for abstinence
No significant increase in sustained abstinence
(Hannöver et al. 2008)

162 sessions in treatment-arm recorded


Transcribed by professional transcription service
Coded by six coders
Sequential Analyses, ICMI 2014
Coding

German Translation of MI-SCOPE










40 hrs. face to face training by first author
Translation of manual + training material
Training of two coders till interobserver agreement consistently
high
Double parsing and double coding of 20 % random samples each
Assessment of interobserver agreement on whole transcripts
Kappa for parsing between
.70 and .96
Kappa for coding between
.50 and .82
Likewise training of three more coders
Ongoing double coding with 5 – 10 minutes random samples of
transcripts
Multicoder kappas between .62 and .79
Sequential Analyses, ICMI 2014
Statistical Analyses

Sequential Analysis




Investigates the occurence of an event, given another event
preceded this event
May be expressed in the form of conditional probabilities;
i.e. the probability p of event E given event G preceded E; p(E | G)
May also be estimated as odd ratios OR and their confidence
intervals
if 95% confidence interval does not include 1 -> statistical
significance may be inferred
OR ≥ 3.0 or OR ≤ 0.33 indicate strong relationships
(Bakeman & Quera, 2011)
Aggregated across observations, contingency tables may be
computed, cell deviations from the expeted frequencies be
computed and tested for statistical significance via their chi-square
distribution
Sequential Analyses, ICMI 2014
Statistical Analyses

Aggregated variables from coded transcripts





MICO:
affirmations, emphasizing control, permission seeking, support
MIIN:
advising (without permission), confronting, directing, stating an
opinion disguised as fact to persuade, warning
CT:
DARN CTs + other – directed towards change
ST:
DARN CTs + other directed towards maintaining behavior
Neutral:
asking, follow/neutral
Sequential Analyses, ICMI 2014
Results overall




Altogether 162 sessions
Overall 78 299 observations
 MICO:
933
(1 %)
 MIIN:
411
(1 %)
 CT:
7 220 (9 %)
 ST:
4 742 (6 %)
 Neutral: 16 553 (21 %)
 Don‘t add to 100 %;
coded utterances like facilitators („hm“ … 33%)
Aggregating and focusing on five categories drastically reduces
frequencies
total number of observations:
659 (0.8 %)
Testing the two technical hypotheses
 MICO:
434
(66%)
 MIIN:
225
(24 %)
 CT:
123
(19 %)
 ST:
80
(12 %)
 Neutral: 456
(69 %)
Sequential Analyses, ICMI 2014
Results MICO / MIIN

Frequencies (Conditional probabilities)
MICO
MIIN
CT
95 (.22)
28 (.12)
ST
32 (.07)
48 (.21)
neutral
307 (.71)
149 (.66)
Χ2 = 31.31; df = 2; p ≤ .01, G2 = 30,23; df = 2; p ≤ .01

Odds ratios (95% CI)
MICO
MIIN
CT
1.97 [1.25 – 3.11]
0.51 [0.32 – 0.80]
ST
0,29 [0.18 – 0.47]
3.41 [2.11 – 5.51]
Sequential Analyses, ICMI 2014
neutral
1.23 [0.87 – 1.74]
0.81 [0.57 – 1.15]
Results / MICO MIIN


MICO increases probability for CT significantly
MICO decreases probability for ST significantly
strong effect

MIIN increases probability for ST significantly
strong effect
MIIN decreases probability CT significantly

Neither MICO nor MIIN significantly predict neutral utterances

Sequential Analyses, ICMI 2014
Results Questions
Frequencies (Conditional probabilities)
Questions
CT
ST
negative
256 (.53)
60 (.12)
positive
79 (.24)
135 (.40)
neutral
992 (.19)
574 (.40)

neutral
170 (.71)
122 (.36)
3 546 (.69)
Χ2 = 549.38; df = 4; p ≤ .01, G2 = 446.69; df = 4; p ≤ .01

Odds ratios (95% CI)
Questions
negative
positive
neutral
CT
4.55 [3.76 – 5.50]
1.07 [0.83 – 1.39]
0.35 [0.30 – 0.41]
ST
neutral
0.94 [0.71 – 1.25] 0.26 [0.21 – 0.32]
5.26 [4.16 – 6.64] 0.29 [0.23 – 0.36]
0.41 [0.34 – 0.49] 4.11 [3.52 – 4.80]
Sequential Analyses, ICMI 2014
Results Questions

Asking for negative aspects of traget behavior



Asking for positive aspects of target behavior



significantly increases probability for CT (strong effect)
significantly decrases probability for neutral utterances (strong
effect)
significantly increases probability for ST (strong effect)
significantly decreases probability for neutral utterances
Asking for neutral information


significantly decreases probability for CT and ST
significantly increases probability for neutral utterances
(strong effect)
Sequential Analyses, ICMI 2014
Results Reflections
Frequencies (Conditional probabilities)
Reflections
CT
ST
positive
599 (.30)
107 (.05)
negative
109 (.10)
289 (.26)
double sided 28 (.15)
42 (.22)
neutral
124 (.05)
111 (.05)

neutral
1 265 (.64)
720 (.64)
118 (.63)
4 149 (.90)
Χ2 = 988.89; df = 6; p ≤ .01, G2 = 901.41; df = 6; p ≤ .01

Odds ratios (95% CI)
Reflections
positive
negative
double s.
neutral
CT
5.56 [4.75 – 6.52]
0.53 [0.43 – 0.66]
0.95 [0.63 – 1.44]
0.20 [0.16 – 0.24]
ST
0.41 [0.33 – 0.51]
5.60 [4.67 – 6.73]
2.76 [1.93 – 3.94]
0.33 [0.27 – 0.41]
Sequential Analyses, ICMI 2014
neutral
0.44 [0.39 – 0.49]
0.53 [0.46 – 0.61]
0.56 [0.41 – 0.76]
4.86 [4.17 – 5.66]
Results Reflections

Offering a positive reflection



Offering a negative reflection



significantly increases probability for ST (strong effect)
significantly decreases probability for CT and neutral utterances
Offering a double-sided reflection



significantly increases probability for CT (strong effect)
significantly decrases probability for ST or neutral utterances
significantly increases probability for ST (strong effect)
significantly decreases probability for neutral utterances
Offering a neutral reflection


significantly increases probability for neutral utterances
(strong effect)
Significantly decreases both CT and ST
Sequential Analyses, ICMI 2014
Discussion

Technical Hypotheses of MI could not be rejected with our data


Path MIIN -> ST strong effect
Hypotheses about questions and reflections also could not be
rejected with our data

Strong effects for following paths:
negative aspects of behavior
-> CT
positive aspects of behavior
-> ST
neutral aspects
-> neutral language
positive reflections
negative reflections
neutral reflection
-> CT
-> ST
-> neutral language
Sequential Analyses, ICMI 2014
Discussion




Few observations of MICO and MIIN despite the big sample
 Especially MIIN
 Emphasis on exploration and reflection
Neutral utterances by far the most frequent category
 Large portion of session dedicated to smoking history and child
protection
First time analyses of questions and reflections
 Processes inside sessions seem to operate as expected
 Large numbers complicate statistical inference – falling back on
guideline for effect size interpretation
Processes from operant conditioning may help in theoretical
formulations for processes in MI
Sequential Analyses, ICMI 2014
Implications for Practice

Using MIIN will more likely result in ST
Using more MICO will more likely result in CT
If you wish to hear CT
 Ask for it – especially the negative aspects of the target behavior
 Reflect it

Implications for training





very easy and straightforward guidance
Medical students find this very helpful
Emphasize training skills for hearing DARN CATs
Supported by German Research Foundation (DFG) Grant No.: HA
5516 / 3-2
Sequential Analyses, ICMI 2014