The current issue and full text archive of this journal is available at www.emeraldinsight.com/0263-4503.htm
A decision tree approach to modeling the private label apparel consumer Marguerite Moore
A decision tree approach
59
Department of Textile and Apparel Technology Management, College of Textiles, North Carolina State University, Raleigh, North Carolina, USA, and
Jason M. Carpenter Department of Retailing, University of South Carolina, Columbia, South Carolina, USA Abstract Purpose – The purpose of this paper is to profile the private label apparel consumer using demographic and behavioral predictors. The paper also aims to examine cross-shopping behaviors among purchasers of private label apparel across the five top US private label apparel retailers. Design/methodology/approach – Decision tree analysis is used to model the impacts of demographics and behaviors on private label purchasing. A secondary database (n ¼ 1,289) of US private label purchasers provides data for the analysis. Findings – Findings indicate demographic predictors as important drivers of private label apparel purchase among retailers positioned as providers of value, while behavioral drivers are more common among patrons of retailers that are differentiated on service or brand. Cross-shopping is more common among the retailers positioned on value. Research limitations/implications – The research design provides a profile of the private label consumer but does not explain why this consumer chooses private labels over national brands. The data-mining approach provides an innovative tool for identifying the drivers of private label consumption. Future research should investigate these drivers more deeply, to establish a fuller understanding of this consumer. The sample is limited to US consumers. Practical implications – Findings suggest that retailers positioned on value/low price need to differentiate private labels to deter cross-shopping. Likewise, comparatively upscale retailers need to continue to be sensitive to the behavioral demands of their respective target market. Originality/value – Results provide a profile of the private label consumer and offer insight into private label cross-shopping using an innovative modeling approach that facilitates examination of actual purchase data. Keywords Decision trees, Labelling, Consumer behaviour, United States of America Paper type Research paper
Introduction Given the global economic downturn of 2008-2009, already tough retail markets have become increasingly challenging for incumbent firms that must fight for decreasing consumer demand. Prior to this economic contraction, the US retail industry was experiencing growth in private label sales across product categories as diverse as grocery, apparel and footwear and home furnishings. In 2006, private label accounted for 45 percent of total US apparel sales. During the same year, two US companies lead global growth of private label sales: Wal-Mart with $126 billion and Target with
Marketing Intelligence & Planning Vol. 28 No. 1, 2010 pp. 59-69 q Emerald Group Publishing Limited 0263-4503 DOI 10.1108/02634501011014615
MIP 28,1
60
$17 billion (Lincoln and Thomassen, 2008). The strategic benefits of private label integration have been noted by both practitioners and empirical researchers (Ailawadi et al., 2008; Tarzijan, 2004). Academic research into private label adoption by consumers focuses primarily upon the grocery industry in markets outside of the USA. Despite the significant resources that US retailers have committed to private label programs, we know very little about the contemporary profile of the private label consumer. The purpose of this study is to provide a demographic and behavioral profile of the private label consumer and to investigate their cross-shopping behaviors using actual purchase data. The study is designed to provide retailers with realistic, actionable directions to effectively target the private label consumer within their strategic approach. Literature The private label consumer Lincoln and Thomassen (2008, p. 6) define private label simply as, “brands owned and sold by the retailer and distributed by the retailer.” As the practice of private label development increased over the years, academics began to investigate different aspects of its effectiveness from supply-chain, firm and consumer perspectives. Within this body of work, consumer research focused primarily upon the interplay between store loyalty and private label choice (Ailawadi et al., 2008; Corstjens and Lal, 2000; Sudhir and Talukdar, 2004; Richardson et al., 1996; Steenkamp and Dekimpe, 1997) and the consideration of national versus private label brands (Ailawadi et al., 2001; Batra and Sinha, 2000). Findings within this stream of research tend to be mixed and do not offer a consensus on the drivers of consumers’ private label choice decision, though researchers consistently suggest that store loyalty likely plays a role in private label purchasing (Richardson et al., 1996). Ailawadi et al. (2008) noted in a recent study that the relationship between store loyalty and private label choice remains in question. They undertook a study of Dutch grocers to address this gap and found that greater private label sales for a retailer drive greater sales overall, however private label customers are often more loyal to low prices than to a particular retailer. The literature tends to focus more heavily on private label in the grocery industry and cultures outside of the USA (Shannon and Mandhachitara, 2005). Cross-shopping behavior Based in part upon Ailawadi et al. (2008) findings, we also examine cross-shopping behaviors among the five formats to investigate whether consumers are exclusive to one retailer when buying private label or split their purchases among multiple private label providers. For the purposes of this study, cross-shopping refers to the same consumer shopping multiple outlets for substitutable private label products. The cross-shopping literature is very limited with only a few examples of empirical inquiry into apparel cross-shopping (Cassill and Williamson, 1994; Carpenter and Moore, 2009). To date, there is no empirical work in the area of cross-shopping for either national brands or private label. Methodology The most effective tool for examining the impact of both behavioral and demographic variables on private label purchasing, in a simultaneous manner, is the relatively new
approach to data analysis that employs classification trees, referred to as decision tree modeling. Decision tree analysis evolved in the field of data mining and is most useful for approaching problems that can be addressed through extensive and diverse data (Garver, 2002). The method allows the researcher to predict a dependent target variable by incorporating numerous and diversely measured independent variables into the model. In many cases, the flexibility and reduced operational constraints of decision tree modeling deem it preferable to logistic regression. Data for the study are provided by the retail forward shopperscape database which uses an online consumer panel to survey shopping behavior across an exhaustive variety of retail formats and product categories. The panel includes approximately one million households across the USA. Participants are recruited through different channels including web portals, web communities, web aggregators, and internet advertising firms. Demographic data are weighted to reflect the US census. All respondents designate themselves as the primary household shopper. Data are collected on a monthly basis with an average response rate of 46 percent (www.retai lforward.com). The data for this study were collected during July 2007 and focused upon consumers’ private label purchasing and shopping behavior across five distinct retailers: Wal-Mart, Target, Kohl’s, JC Penney, and Macy’s. The five retailers selected for the study are recognized as leaders in the US private label apparel market. The retailers vary somewhat in terms of format and/or strategic approach. Wal-Mart and Target are both considered discount formats, though Wal-Mart is recognized for low prices and Target for more fashionable apparel brands. Kohl’s is considered a value department store that is highly promotional and also noted for low prices. JC Penney and Macys represent the traditional department store format but offer distinct retail mixes, with JC Penney positioned as a value retailer and Macy’s positioned as more upscale, offering comparatively higher prices. Based upon demonstrated success over time, specific private label brands were selected to represent each retailer to model purchase behavior: Faded Glory in Wal-Mart, Merona in Target, Sonoma in Kohl’s stores, Arizona in JC Penney and Charter Club in Macy’s. The dependent/target variable captures actual purchase behavior and is represented by a binomial measure for private label purchase within each of the five retailers (1 – yes, 0 – no). Again, each purchase variable is associated with a specific private label brand: Faded Glory in Wal-Mart, Merona in Target, Sonoma in Kohl’s stores, Arizona in JC Penney and Charter Club in Macy’s. In order to model actual purchase behavior, the dependent variables were associated with specific purchases of the identified private label products. The predictors of private label purchasing include both demographic variables and shopping behaviors. Demographic variables include: age, income, education level, household size, ethnicity, and marital status. Shopping behavior variables include shopping frequency on a three-point scale from never-frequent (mall, discount center, online) and propensity to bargain shop or to splurge (1 – yes, 0 – no). Additional decision trees are performed on the target variable for each retail format using purchase and shopping behaviors from the remaining four formats as the predictors to determine cross-shopping behavior. All variables are measured on ordinal or nominal scales. Chi-square Automatic Interaction Detector (CHAID) is used to model the predictors of private label purchase behavior due to the nominal nature of the target variables.
A decision tree approach
61
MIP 28,1
62
CHAID uses x 2 and F-tests to select variables and split them into groups referred to as nodes. CHAID is capable of splitting significant variables into multiple nodes. A prori model settings include: a maximum of five levels for the tree, a minimum of 60 cases per parent node and 25 cases per child node. Pearson x 2 are used to detect differences in the cases with alpha designated at 0.01 for both splitting and merging. Of the sample (n ¼ 860), 65 percent is used for the initial model, commonly termed as the training model while the remaining 35 percent (n ¼ 429) of the sample is used as the testing or validation model. In cases that sample sizes are limited, the training model is usually tested on a larger proportion of the data (SPSS, 2002). Results CHAID generated five distinct models associated with private label purchasing in each of the retailers (Figures 1-5). Three additional models that examined cross-shopping among the retail formats were also generated (Figures 6-8). All eight models exhibited identical structure and similar risk estimates between respective training and testing samples (Table I), which suggests that each model generalizes well across the data. Based upon the level of agreement between the training and test models, interpretation of the decision trees (training models) proceeded. The five trees designed to examine the demographic and behavioral dimensions of the private label consumer indicate somewhat distinct profiles among the retailers examined. Within the decision tree model, the order of variable splits from top to bottom indicates the magnitude of effects in terms of influence on the target variable. Two of the five models exclusively generated demographic predictors (Target and Kohl’s) while the remaining three models generated significant demographic and behavioral predictors. Y = 28% N = 72% Split 1: Household size X2 = 22.65, p< 0.003 3-4, 6 members 1-2,7 members
Split 2: Income X2 = 8.39, p< 0.011
Y = 22% N = 78%
< $49K
$49K > Y = 28% N = 72% Married
Split 3: Marital status X2 = 6.73, p< 0.028 Note: n = 860
Y = 48% N = 52%
Y = 33% N = 67%
Y = 45% N = 55%
Figure 1. Wal-Mart private label purchasers demographics and behavior
5 members
Y = 31% N = 69%
Not married
Y = 9% N = 91%
A decision tree approach
Y = 11% N = 89%
Split 1: Bargain shopper X2 = 6.56, p 45
Y = 11% N = 89%
63
Y = 3% N = 97% Occasional/ infrequent
Frequent Y = 19% N = 81%
Split 2: Age X2 = 9.661, p< 0.009
Y = 9% N = 91%
Split 3: Discount shopping frequency X2 = 6.36, p< 0.035
Note: n = 860
Figure 2. Target private label purchasers demographics and behavior
Y = 17% N = 83%
Split 1: Income X2 = 16.85, p< 0.0007 < $25K Split 3: Age X2 = 7.65, p< 0.0283 < 44 Y = 3% N = 97%
$25K-$99K
Y = 9% N = 91%
$99K >
Y = 16% N = 84% 44 >
Y = 16% N = 84%
Y = 26% N = 74% < 43 Y = 23% N = 77%
Split 2: Age X2 = 9.949, p< 0.0145 No 43 > Y = 13% N = 87%
Note: n = 860
Household size (x 2 ¼ 22.65, p , 0.003) was the most influential variable for predicting private label purchase among Wal-Mart consumers, followed by income (x 2 ¼ 8.39, p , 0.011) and marital status (x 2 ¼ 6.73, p , 0.028) (Figure 1). For the most part, purchases of private label increased as household size increased. The second split indicated that among medium-sized households (three to four, six members) private label purchasers tend to belong to a lower income group ($25,000-49,999). Among the higher income group ($49,999 . ), purchasers were much more likely to be married than single. The second model which investigated private label purchasing in Target generated three levels of significant predictors (Figure 2). The propensity to bargain shop (x 2 ¼ 6.56, p , 0.010) generated the first split followed by age (x 2 ¼ 9.66, p , 0.009) and frequency of discounter shopping (x 2 ¼ 6.36, p , 0.035). A greater proportion of private label purchasers at Target tended not to bargain shop. The smaller group of purchasers that indicated a propensity to bargain shop tended to be younger (, 45) and frequent shoppers of discount formats. The third model examined the drivers of private label purchase among the Kohl’s group (Figure 3). The model indicated two significant predictors: income and age.
Figure 3. Kohl’s private label purchasers demographics and behavior
MIP 28,1
Y = 21% N = 79% Split 1: Household size X2 = 34.43, p = 0.000 1-2 members
64
Y = 14% N = 86%
Split 3: Age X2 = 16.54, p< 0.009 < 43 Split 5: Mall shopping frequency X2 = 10.68, p< 0.003 Occasional/never
Figure 4. JC Penney private label purchasers demographics and behavior
Y = 18% N = 82% Some college >
< Some college
Y = 23% N = 77%
Y = 45% N = 55%
Y = 36% N = 64%
Y = 42% N = 58%
Some college > Y = 21% N = 79% Frequent/ never
Occasional
Y = 9% N = 91%
Y = 30% N = 70%
Y = 14% N = 86%
Split 6: Online shopping frequency X2 = 7.56, p< 0.017
Split 4: Education X2 = 11.63, p< 0.005
Note: n = 860
Split 1: Shopping frequencymall X2 = 13.34, p< 0.0008
Y = 6% N = 94% Occasional/never
Y = 5% N = 95% Split 2: Race X2 = 13.43, p< 0.0007
Caucasian, African American
$99K >
Y = 3% N = 97% Occasional
Note: n = 860
Y = 8% N = 92% Never
Y = 4% N = 96%
Y = 16% N = 84%
Y = 4% N = 96%
< $50-$99K
Figure 5. Macy’s private label purchasers demographics and behavior
Split 2: Education X2 = 19.40, p< 0.001
< HS grad < Some college
55 > Y = 10% N = 90%
Y = 24% N = 76%
Frequent
Y = 4% N = 96%
Y = 27% N = 73%
43-55
Y = 9% N = 91%
6-7 members
3-5 members
Y = 0% N = 100%
Frequent
Y = 12% N = 88%
Hispanic, Asian, other
Split 3: Income X2 = 8.62, p< 0.009
Split 4: Shopping frequency-mall X2 = 4.47, p< 0.035
A decision tree approach
Y = 28% N = 72% Yes
Split 1: Purchased JC Penney X2 = 78.35, p = 0.000
No
Y = 55% N = 45%
Y = 21% N = 79%
Split 2: Purchased Kohl’s X2 = 12.64, p< 0.0004
65
No
Yes Y = 34% N = 66%
Y = 81% N = 19%
Figure 6. Wal-Mart private label purchasers cross-shopping
Note: n = 860
Y = 17% N = 83% Split 1: Purchased Wal-Mart X2 = 9.49, p< 0.002
Yes
No
Y = 88% N = 12%
Y = 69% N = 31%
Note: n = 680
Split 1: Purchased Wal-Mart X2 = 78.35, p = 0.000
Y = 21% N = 79%
Yes
No
Y = 40% N = 60% Yes Y = 60% N = 40%
Figure 7. Kohl’s private label purchasers cross-shopping
Y = 13% N = 87% No Y = 36% N = 64%
Split 2: Purchased target X2 = 7.91, p< 0.004
Note: n = 860
Income (x 2 ¼ 16.85, p , 0.0007) indicated the greatest influence on private label purchasing followed by age which split twice beneath separate income groups. The greatest proportion of private label purchasers were from moderate ($25,000-99,999) and high ($99,999 . ) income groups. The model indicated the second split with the
Figure 8. JC Penney private label purchasers cross-shopping
MIP 28,1
66 Table I. Risk estimates for decision trees
Risk estimate Model 1. 2. 3. 4. 5. 6. 7. 8.
Wal-Mart: demographics and behavior Target: demographics and behavior Kohl’s: demographics and behavior JC Penney: demographics and behavior Macy’s: demographics and behavior Wal-Mart: cross-shopping Kohl’s: cross-shopping JC Penney: cross-shopping
Traininga
Testingb
0.2813 0.1116 0.1697 0.2058 0.0593 0.2616 0.0534 0.1965
0.3006 0.1072 0.1911 0.2144 0.0769 0.2867 0.0442 0.2074
Notes: an ¼ 860; bn ¼ 429
high-income group in terms of age (x 2 ¼ 9.94, p , 0.014). A greater proportion of the moderate income buyers were younger (, 43). The third split, again associated with age (x 2 ¼ 7.65, p , 0.0283), suggested that the larger group of low income private label purchasers at Kohl’s were slightly older (44 . ). The fourth model indicated six significant predictors for private label purchase among the JC Penney group: household size (x 2 ¼ 34.43, p ¼ 0.000), education (x 2 ¼ 19.40, p , 0.001), age (x 2 ¼ 16.54, p , 0.009), education (x 2 ¼ 11.63, p , 0.005), mall shopping frequency (x 2 ¼ 10.68, p , 0.003), and online shopping frequency (x 2 ¼ 7.56, p , 0.017) (Figure 4). The strongest predictor, household size, increased proportionately with the occurrence of private label purchasing at JC Penney. Households with three to five and six to seven members were more likely to purchase private label than smaller households (one to two members). Purchasers among the three to five member household group varied significantly in terms of education. The greatest proportion of this group was moderately educated with a high school degree/some college. A smaller group of college educated purchasers indicated occasional online shopping behavior. Among the smaller households, middle aged (43-55) respondents tended to purchase more private label at JC Penney compared to younger (,43) and older (55.) groups. This group of purchasers also indicated some college as the highest level of education. Younger private label purchasers at JC Penney tended to be frequent mall shoppers. The model for the Macy’s purchasers indicated three significant predictors: mall shopping frequency, race (x 2 ¼ 13.43, p , 0.0007) and income (x 2 ¼ 8.62, p , 0.009) (Figure 5). Mall shopping frequency indicated effects at the first (x 2 ¼ 13.34, p , 0.0008) and fourth (x 2 ¼ 4.47, p , 0.035) level of the tree. A larger proportion of Macy’s private label purchasers reported that they were frequent mall shoppers. Among the smaller group who reported that they only occasionally shopped in malls, the majority of customers had incomes above $99,999 per annum. An additional series of decision trees were also generated to identify whether the private label customer was cross-shopping within retailers in the sample (Figures 6-8). Each original target variable for each retailer (purchased private label apparel (1 – yes, 0 – no)) was modeled using the private label purchase variables from the remaining four retailers as predictors to determine cross-shopping behaviors. In addition, four
variables that simply asked whether consumers had shopped in each retailer for private labels (1 – yes, 0 – no) were also examined as predictors within these models. Three of the five decision trees indicated significant cross-shopping effects: Wal-Mart, Kohl’s, and JC Penney. The Target and Macy’s private label purchasers did not indicate shopping or purchasing private label apparel from the other retailers. Both Kohl’s and JC Penney purchasers indicated significant private label purchases from Wal-Mart. Vice-versa, the Wal-Mart model indicated the same effects with JC Penney accounting for heavier private label cross-shopping followed by Kohl’s. The model for JC Penney also indicated that their private label purchasers also purchase private label from Target. Conclusions, implications, and future research The decision trees yielded varying models to describe the private label consumer. Demographic indicators were more prominent than behavioral characteristics when predicting private label purchase across the five retailers. In particular, household size and income represented the primary predictor for private label purchasing in formats that are positioned on price (Wal-Mart, Kohl’s, and JC Penney). Both Target and Macy’s private label purchasers indicated behavioral predictors as significant drivers of choice. Target private label patrons indicated low levels of bargain shopping while Macy’s patrons were frequent mall shoppers. Findings suggest that private label purchasers are more different than similar in terms of demographics and behaviors according to the selected retail format, with several exceptions between the Wal-Mart, Kohl’s and JC Penney formats which appear to share competitive space. The cross-shopping trees indicated that Wal-Mart and JC Penney private label consumers engage in purchasing comparable private labels in both chains. Wal-Mart and Kohl’s also share private label consumers for comparable private label brands. Further, JC Penney private label purchasers patronize Target for private label apparel. The higher incidence of cross-shopping among Wal-Mart, Kohl’s, and JC Penney could be attributed to the positioning strategies of this group as high-value/low-price providers within their respective channels of distribution. Respondents did not indicate significant levels of cross-shopping for private label among the Target or Macy’s group, suggesting that these retailers may have higher competitive barriers for their private label brands compared to the other retailers in the study. In conclusion, the findings reinforce the logic that retailers should position private label programs to accurately reflect the core value proposition within their respective marketing strategies. Further, the findings suggest that retailers facing tough experiencing higher incidences of private label cross-shopping in the US market (Wal-Mart, Kohl’s, and JC Penney) could benefit from differentiating their private labels within their offering, thereby encouraging trading up behaviors, potentially higher margins and ultimately greater loyalty. This study provides a superficial examination of the characteristics of the private label consumer. The design is focused upon what rather than why in explaining the private label purchase choice. The study sacrifices deep understanding for a rigorous, comprehensive overview of the practical drivers of consumer behavior in private label purchasing. Additional behavioral variables including a more comprehensive conception of shopper orientation may have resulted in more robust models. Further, we have no knowledge whether our subjects recognize private labels as distinct from other types of brands, or their degree of loyalty to the retailer of choice.
A decision tree approach
67
MIP 28,1
68
From a methodological standpoint, larger datasets with longitudinal data would have strengthened the credibility of the trees by cross-validating the models over time and different consumer samples. However, the variables considered in our models are commonly used by retailers to segment markets and position brands due to their accessibility and utility in representing homogeneous groups of consumers. In keeping with directions already suggested in the extant literature, future research within this stream should consider the interaction of private label brands with national brands to more fully understand the drivers of brand choice in the retail context and ultimately inform merchandising and marketing strategy. Further, findings suggest that private label consumer profiles were fairly well aligned with the target consumers of each retailer, which suggests that these consumers may exhibit loyalty towards the retailer and purchase its private labels as a result of that loyalty. Again, this question remains unanswered in the extant literature and provides a potentially important direction for understanding the private label consumer within the apparel context.
References Ailawadi, K.L., Neslin, S. and Gedenk, K. (2001), “Pursing the value conscious consumer: store brands versus national brand promotions”, Journal of Marketing, Vol. 65, pp. 71-89. Ailawadi, K.L., Pauwels, K. and Steenkamp, J.B. (2008), “Private label users and store loyalty”, Journal of Marketing, Vol. 72 No. 6, pp. 19-30. Batra, R. and Sinha, I. (2000), “Consumer level factors moderating the success of private label brands”, Journal of Retailing, Vol. 76 No. 2, pp. 175-91. Carpenter, J. and Moore, M. (2009), “Consumer demographics, retail attributes, and apparel cross shopping behavior”, Journal of Textile and Apparel Technology and Management, Vol. 6 No. 1, pp. 1-14. Cassill, N. and Williamson, N. (1994), “Department store cross-shoppers”, Journal of Applied Business Research, Vol. 10 No. 4, pp. 88-97. Corstjens, M. and Lal, R. (2000), “Building store loyalty through store brands”, Journal of Marketing Research, Vol. 37, pp. 281-92. Garver, M. (2002), “Using data mining for customer satisfaction research”, Marketing Research, Vol. 14 No. 1, pp. 8-12. Lincoln, K. and Thomassen, L. (2008), Private Label: Turning the Retail Brand into Your Biggest Opportunity, Kogan Page, London. Richardson, P.S., Jain, A. and Dick, A. (1996), “Household store brand proneness: a framework”, Journal of Retailing, Vol. 72 No. 2, pp. 159-85. Shannon, R. and Mandhachitara, R. (2005), “Private-label grocery attitudes and behavior: a cross-cultural study”, Journal of Product & Brand Management, Vol. 12 No. 6, pp. 461-74. SPSS (2002), Introduction to Answer Tree, SPSS, Chicago, IL. Steenkamp, J.B. and Dekimpe, M.G. (1997), “The increasing power of store brands: building loyalty and market share”, Long Range Planning, Vol. 30 No. 6, pp. 917-30. Sudhir, K. and Talukdar, D. (2004), “Does store brand patronage improve store patronage?”, Review of Industrial Organization, Vol. 24, pp. 143-60. Tarzijan, J. (2004), “Strategic effects of private labels and horizontal integration”, International Review of Retail Distribution and Consumer Research, Vol. 14 No. 3, pp. 321-35.
About the authors Marguerite Moore is an Associate Professor at North Carolina State University. Her research and teaching is primarily in the areas of retail strategy, brand management and consumer behavior in textiles. Marguerite Moore is the corresponding author and can be contacted at:
[email protected] Jason M. Carpenter, PhD, is an Assistant Professor of Retailing at The University of South Carolina. His work has been published in the Journal of Retailing & Consumer Services, International Journal of Retail & Distribution Management, and the Journal of Product & Brand Management.
To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
A decision tree approach
69