Skip to main content

Eliciting brand-related social media engagement: A conditional inference tree framework

Journal of Business Research, 2021

Bruno Schivinski

Journal of Business Research, volume 130, pages 594–602, 2021 (online September 2019)

  • 1,356 social media users
  • 282 brands
  • Machine learning (conditional inference tree)
  • Poland
Download the accepted manuscript (PDF)Published version

Free full text: the author's accepted manuscript, identical in content to the published article and shared under CC BY-NC-ND 4.0. Please cite the published version.

Keywords
  • CBBE
  • COBRAs
  • social media
  • machine learning
  • typology
  • consumer brand engagement
  • brand-related social media engagement
  • behavioural engagement
  • consumers' online brand-related activities
  • content consumption, contribution and creation
  • user-generated content
  • brand commitment
  • brand loyalty
  • brand awareness
  • brand recognition
  • brand familiarity
  • brand associations
  • consumer segmentation
  • consumer typology
  • brand fans
  • consumer-based brand equity
  • Aaker brand equity model
  • hierarchy of effects
  • theory of reasoned action
  • CEBSC scale
  • Poland
  • social media users
  • conditional inference tree (Ctree)
  • decision tree
  • recursive partitioning
  • partykit
  • predictive modelling
  • confirmatory factor analysis
  • survey research
  • social media strategy
  • brand management
  • customer engagement marketing
  • audience segmentation

Four things consumers think about a brand, how committed they are to it, whether they recognise it, how familiar it is and whether its associations are favourable, sort social media users into five engagement patterns. A machine learning tree on 1,356 users following 282 brands in Poland showed that brand commitment is the first dividing line.

Key findings

  1. Four brand equity items predict who engages with brand content. In a survey of 1,356 social media users in Poland, each rating a brand they followed (282 brands), a conditional inference tree selected four of twelve brand equity items as predictors of consuming, contributing to and creating brand-related social media content: brand commitment, brand recognition, brand familiarity and favourable brand associations (Schivinski, 2021).
  2. Brand commitment is the first split. The tree first divided consumers at a score of 4 on a 1–7 agreement scale for “I am committed to this brand” (p < .001); committed consumers were then split by brand familiarity and less committed ones by brand recognition and favourable associations (Schivinski, 2021).
  3. Most consumers are brand passionates. The largest group, the brand passionate (rule 1, n = 625), combined average or low commitment (≤ 4), high brand recognition (> 4) and favourable associations above 3; most of them consumed (87.15%) and contributed to (82.44%) brand content at a medium level, while 54.13% created little content (Schivinski, 2021).
  4. Brand fans are the most active creators. Brand fans (rule 2, n = 558), with high commitment and high familiarity (both > 4), had the largest shares of high engagement: 21.17% contributed and 17.96% created brand content at a high level (Schivinski, 2021).
  5. Weak associations hold engagement down. Low engaged consumers (rule 3, n = 108) recognised the brand (> 4) but rated its associations as unfavourable (≤ 3); 65.9% created little or no brand content and 42.59% contributed at a low level (Schivinski, 2021).
  6. Loyal consumers who know the brand less well create more. Prospect brand fans (rule 5, n = 24) were committed to the brand (> 4) but less familiar with it (≤ 4); 34% of them created brand content at a high level (Schivinski, 2021).

At a glance

Design
Cross-sectional online survey (Qualtrics); confirmatory factor analysis (lavaan, robust maximum likelihood) and a conditional inference tree (partykit) in R
Participants
1,593 social media users recruited through brand pages in Poland; 1,356 analysed after removing incomplete cases and outliers (56.7% women; mean age 29, SD 8.5, range 18–52; 97.78% used smartphones for social media)
Brands
282 brands across industries, including beverages, cars, clothing, fashion and beauty, retail, hospitality and sporting goods; each respondent rated a brand they follow on social media
Measures
Consumer-based brand equity: brand awareness, brand associations, perceived quality and brand loyalty (12 items; Christodoulides et al., 2015; 1–7 agreement). Consumption, contribution and creation of brand-related social media content (17 items, CEBSC scale; Schivinski et al., 2016; 1–7 frequency), split into low, medium and high at the upper and lower deciles
Model
Conditional inference tree with 95% confidence, minimum split of 70 cases, controlling for age, gender, social media usage and brands; five terminal nodes (rules)
Model fit
CFA: MLR χ²(356) = 1343.12, CFI = 0.92, TLI = 0.91, RMSEA = 0.06 [90% CI 0.06–0.07], SRMR = 0.06; α and CR 0.81–0.92; AVE 0.59–0.75

Figure 1. Conditional inference tree predicting brand-related social media engagement

1. Brand commitment[BL3] I am committed to this brand2. Brand recognition[BAW3] I can recognize this brand…7. Brand familiarity[BAW2] I am quite familiar with this brand4. Favourable associations[BAS2] This brand has favorable associations≤ 4> 4≤ 4> 4≤ 3> 3≤ 4> 4Node 3 · Rule 4Brand curiousn = 41ConsumptionContributionCreationNode 5 · Rule 3Low engagedn = 108ConsumptionContributionCreationNode 6 · Rule 1Brand passionaten = 625ConsumptionContributionCreationNode 8 · Rule 5Prospect brand fann = 24ConsumptionContributionCreationNode 9 · Rule 2Brand fann = 558ConsumptionContributionCreationLow engagementMedium engagementHigh engagementBars show the share of each node at each level (0–100%). Scales 1–7.
Each split uses one brand equity item on a 1–7 agreement scale (> 4 means endorsing the item at 5 or above). Terminal nodes show the share of consumers with low, medium and high consumption, contribution and creation of brand-related social media content. Minimum split = 70; controls: age, gender, social media usage and brands.

Share of consumers in each engagement pattern

Brand passionate (rule 1)46.1%Brand fan (rule 2)41.2%Low engaged (rule 3)8%Brand curious (rule 4)3%Prospect brand fan (rule 5)1.8%
Size of the five terminal nodes of the conditional inference tree as a share of the 1,356 participants (Table 2 of the article).

Tables

Table 2. The five patterns (rules) and subtypes of consumers, contributors and creators of brand-related social media content

Rule Pattern Consumption Contribution Creation Subtype
1 (n = 625, node 6) Brand commitment average/low (≤ 4); brand recognition high (> 4); favourable associations average/high (> 3) Medium 87.15%; low 11.60%; high 1.25% Medium 82.44%; low 15%; high 2.56% Low 54.13%; medium 43.33%; high 2.54% Brand passionate. Moderately consumes and contributes; occasionally creates content
2 (n = 558, node 9) Brand commitment high (> 4); brand familiarity high (> 4) Medium 83.46%; high 11.10%; low 5.44% Medium 73.22%; high 21.17%; low 5.54% Medium 53.52%; low 28.52%; high 17.96% Brand fan. Highly committed and proactive towards COBRAs
3 (n = 108, node 5) Brand commitment average/low (≤ 4); brand recognition high (> 4); favourable associations low (≤ 3) Medium 71.02%; low 27.98%; high 1% Medium 56.11%; low 42.59%; high 1.3% Low 65.9%; medium 34.09% Low engaged. Knows the brand; mainly low and medium engagement
4 (n = 41, node 3) Brand commitment average/low (≤ 4); brand recognition average/low (≤ 4) Medium 91.67%; low 8.33% Medium 93.34%; low 6.66% Split: low 33.32%; medium 33.33%; high 33.34% Brand curious. Engaged with all three COBRAs while commitment and recognition are low
5 (n = 24, node 8) Brand commitment high (> 4); brand familiarity average/low (≤ 4) Medium 82.44%; low 17.56% Medium 66.66%; low 16.9%; high 16.44% Low 50%; high 34%; medium 16% Prospect brand fan. Loyal but lacks breadth of familiarity; prone to higher COBRAs

Scales ranged from 1 to 7 for both CBBE and COBRAs. Low, medium and high engagement were set at the upper and lower deciles (Schivinski et al., 2016).

Table 1. Construct reliability and validity
Construct α CR AVE 1 2 3 4 5 6 7
1. Consumption 0.88 0.88 0.60 0.77
2. Contribution 0.91 0.90 0.62 0.61*** 0.79
3. Creation 0.92 0.92 0.65 0.48*** 0.70*** 0.81
4. Brand awareness 0.81 0.81 0.59 0.04 ns −0.03 ns −0.16*** 0.77
5. Brand associations 0.90 0.90 0.75 0.38*** 0.26*** 0.16*** 0.32*** 0.87
6. Perceived quality 0.83 0.82 0.60 0.22*** 0.20*** 0.16*** 0.33*** 0.75*** 0.77
7. Brand loyalty 0.89 0.89 0.73 0.37*** 0.38*** 0.32*** 0.17*** 0.61*** 0.66*** 0.85

α = Cronbach’s alpha; CR = composite reliability; AVE = average variance extracted. The square roots of the AVE values are in italics. *** p < 0.001; ns = not significant.

Table A1. Scale items: factor loadings, explained variance and descriptive statistics
Item λ R² t Mean SD Skewness Kurtosis
COBRAs framework (Schivinski et al., 2016)
[CONS1] I read posts related to Brand X on social media. 0.86 0.72 37.25 3.52 2.02 0.16 −1.23
[CONS2] I read fan page(s) related to Brand X on social networking sites. 0.86 0.73 36.66 3.54 2.11 0.24 −1.33
[CONS3] I watch pictures/graphics related to Brand X. 0.69 0.48 22.44 3.93 2.05 −0.11 −1.28
[CONS4] I follow blogs related to Brand X. 0.58 0.36 17.13 2.33 1.69 1.13 0.42
[CONS5] I follow Brand X on social networking sites. 0.86 0.77 37.56 3.38 2.01 0.34 −1.12
[CONT1] I comment on videos related to Brand X. 0.81 0.66 16.91 1.87 1.45 1.87 2.88
[CONT2] I comment on posts related to Brand X. 0.86 0.75 21.22 2.03 1.54 1.53 1.86
[CONT3] I comment on pictures/graphics related to Brand X. 0.86 0.73 18.73 1.89 1.49 1.83 2.99
[CONT4] I share Brand X related posts. 0.82 0.72 21.98 2.09 1.51 1.87 1.45
[CONT5] I ‘Like’ pictures/graphics related to Brand X. 0.67 0.42 22.79 3.06 2.01 0.56 −1.11
[CONT6] I ‘Like’ posts related to Brand X. 0.68 0.47 24.25 2.66 2.02 0.68 −0.92
[CREA1] I initiate posts related to Brand X on blogs. 0.70 0.48 13.10 1.74 1.31 2.11 3.72
[CREA2] I initiate posts related to Brand X on social networking sites. 0.79 0.63 15.98 1.79 1.47 2.13 3.41
[CREA3] I post pictures/graphics related to Brand X. 0.88 0.77 19.91 1.83 1.38 1.86 2.95
[CREA4] I post videos that show Brand X. 0.83 0.68 16.54 1.72 1.36 1.89 2.83
[CREA5] I write posts related to Brand X on forums. 0.86 0.76 19.18 1.81 1.41 1.99 2.85
[CREA6] I write reviews related to Brand X. 0.79 0.61 14.57 1.69 1.33 2.11 3.81
CBBE framework (Christodoulides et al., 2015)
[BAW1] I have heard of this brand. 0.69 0.48 9.72 6.70 1.01 −2.54 7.82
[BAW2] I am quite familiar with this brand. 0.78 0.60 11.84 6.62 0.95 −2.21 7.33
[BAW3] I can recognize this brand among other brands. 0.83 0.71 10.11 6.71 0.72 −2.78 6.45
[BAS1] This brand has strong associations. 0.89 0.79 22.92 6.01 1.22 −1.45 1.73
[BAS2] This brand has favorable associations. 0.83 0.70 22.01 6.05 1.45 −1.23 0.91
[BAS3] It is clear what this brand stands for. 0.88 0.77 25.87 5.72 1.55 −1.11 0.65
[PQ1] This brand is good quality. 0.75 0.58 20.45 5.56 1.59 −0.99 0.45
[PQ2] This brand has excellent features. 0.78 0.61 23.04 5.17 1.71 −0.82 −0.23
[PQ3] Compared to other brands in its category, this brand is of very high quality. 0.81 0.65 18.98 5.88 1.55 −1.23 0.61
[BL1] I feel loyal to this brand. 0.90 0.83 35.34 4.58 1.91 −0.42 −0.91
[BL2] This brand is my first choice. 0.89 0.82 33.69 4.57 1.93 −0.44 −1.02
[BL3] I am committed to this brand. 0.77 0.59 26.66 4.25 2.06 −0.04 −1.21

λ = completely standardised loading; R² = explained variance of the item. Scales ranged from 1 to 7 for both CBBE and COBRAs. CONS = consumption, CONT = contribution, CREA = creation; BAW = brand awareness, BAS = brand associations, PQ = perceived quality, BL = brand loyalty.

The question

Brands want people to read, share and create content about them on social media, and research had mapped what engagement is and which personal motives drive it. Less was known about how a consumer's view of the brand itself turns into engagement. I asked which levels of brand equity, item by item, lead consumers to consume, contribute to and create brand-related content, and whether those patterns describe recognisable types of consumers.

Implications

Practical implications

The five patterns give social media managers a way to plan engagement by what consumers think of the brand. For brand passionates, the largest group, managers can build commitment through a shared sense of community on the brand page and keep recognition high by featuring brand extensions and offers. Brand fans respond to personalised content and humanised brand communication. For low engaged consumers, typical of fast-moving goods and business-to-business markets, the first task is to improve brand associations. Brand curious consumers want to learn, so educational and cross-device campaigns suit them, and prospect brand fans, often found among niche and newer brands, can be turned into fans by widening their familiarity with the product range.

Theoretical implications

The study contributes to the behavioural view of consumer brand engagement by showing how specific levels of brand equity translate into consuming, contributing and creating brand content, in line with the hierarchy-of-effects model. Instead of estimating average effects across variables, the conditional inference tree isolates thresholds in individual items and the combinations that lead to each level of engagement, and from those combinations it derives a typology of five consumer subtypes. Brand commitment, from the loyalty dimension, emerged as the most important item.

Limitations and future research

  • The conditional inference tree is exploratory; experiments that manipulate brand equity are needed to confirm the patterns.
  • All respondents already followed the brand on social media, and data came from one country, so the patterns should not be generalised to all consumers or cultures.
  • Differences between brands, industries and high- and low-equity brands were not examined; other brand constructs, such as brand image, attitude or love, could also drive COBRAs.

Questions this paper answers

Which brand perceptions make people engage with brands on social media?

In a machine learning analysis of 1,356 social media users in Poland, four brand equity items predicted engagement: brand commitment, brand recognition, brand familiarity and favourable brand associations, with commitment as the first and most important split (Schivinski, 2021).

What are the five types of brand followers on social media?

Brand passionates (46% of the sample), brand fans (41%), low engaged consumers (8%), brand curious consumers (3%) and prospect brand fans (2%), each defined by a combination of commitment, recognition, familiarity and associations (Schivinski, 2021).

Who creates the most content about brands on social media?

Committed consumers. Among brand fans, highly committed and highly familiar with the brand, 17.96% created brand content at a high level, and among prospect brand fans, committed but less familiar with the brand, 34% did (Schivinski, 2021).

Why do some followers barely engage with a brand they know?

In this study, consumers who recognised the brand but rated its associations as unfavourable (3 or lower on a 1–7 scale) formed the low engaged group; 65.9% of them created little or no brand content (Schivinski, 2021).

Key concepts

Conditional inference tree (Ctree)
A machine learning method that splits a sample step by step on the predictor most strongly associated with the outcome, using permutation tests to avoid selection bias; its results read as if–then rules (Hothorn, Hornik & Zeileis, 2006).
Consumers' online brand-related activities (COBRAs)
Brand-related activities on social media, from passive to active: consuming, contributing to and creating brand-related content (Muntinga, Moorman & Smit, 2011; Schivinski et al., 2016).
Consumer-based brand equity (CBBE)
The perceptions, attitudes, knowledge and behaviour that let a brand earn more volume or margin than it could without its name; measured here through brand awareness, associations, perceived quality and loyalty (Aaker, 1991).
Brand commitment
An item of the brand loyalty dimension (“I am committed to this brand”) and the first split in the tree.
Brand recognition and brand familiarity
Two items of brand awareness: being able to pick the brand out among others, and knowing the brand well.
Rule
A path from the top of the tree to a terminal node: the combination of item thresholds that defines one consumer subtype.

Open science and materials

Study materials

Screening question
Respondents confirmed they had used social media in the past six months and named a brand they follow on social media; that brand was used throughout the survey.
Scale items
All 29 items with loadings and descriptive statistics are in Table A1 below.

How to cite

Schivinski, B. (2021). Eliciting brand-related social media engagement: A conditional inference tree framework. Journal of Business Research, 130, 594–602. https://doi.org/10.1016/j.jbusres.2019.08.045

Related publications

All publications