Composing tweets to increase retweets

Brands tweet not only to communicate with followers but also to reach large audiences rapidly when the tweets are retweeted by the followers. People however will retweet only if they recognize within a few seconds that the tweet is on an interesting topic. Brands therefore need insights into how to compose tweets to facilitate topic recognition even when they are just scanned. This is the issue that we address in this research. Specifically, drawing on findings in psycholinguistics, we empirically investigate if tweets composed such that they include more topic-related words that are located closer to the start get more retweets. Results from an investigation of sales-promotional tweets by sixty brands in four categories indicate that tweets that are composed as above do get more retweets. We repeat the investigation using tweets on several other topics from a natural experiment that generated pairs of tweets where each pair is on the same topic but each tweet in the pair is composed differently. This investigation reconfirms the findings from the analysis of retweets of sales-promotional tweets. We conclude by presenting an approach for how social media managers can compose tweets based on our findings.     

Is Online Word of Mouth Higher for New Models or Redesigns? An Investigation of the Automobile Industry

Changing consumer preferences, entry of new competitors with better products, and product improvements by existing competitors make new products a necessary investment for most companies. In addition to the investments in R&D and production of new products, firms also need to invest in advertising to promote them. The growing use of social media by consumers therefore makes online consumer conversations an attractive additional format for firms to promote products at a lower cost. This is particularly so in the automobile industry where advertising costs are very high. Whether consumers discuss a newly introduced product, and help to promote it, however, may depend on how new the product is. This is the question that we investigate in the context of the automobile category. Specifically, we examine whether online consumer conversations are more likely for new models (e.g., Chevy Volt introduced in 2010) or redesigns of existing models (e.g., the redesign of Chevy Impala in 2010). We use data from two online sites where consumers discuss automobiles – consumerreports.org and edmunds.com – for our analysis. Our empirical investigation also accounts for the effects of sales on word of mouth and the simultaneity between sales and word of mouth. Additionally, we also consider the effect of satisfaction on word of mouth and the endogeneity of satisfaction. Our results across two datasets suggest that redesigns stimulate significantly more conversations than new models. Managerial and research implications of the findings are discussed.     

Shopping Style Segmentation of Consumers

The World Wide Web has rapidly become an alternative means to reach customers and has attracted the attention of many businesses. Unfortunately, however, despite its growth, there is little knowledge of which consumers would be willing to switch to the new format and to what extent. Our paper is aimed at providing some insights into these questions. Specifically, we propose a model to identify segments that differ in their shopping style, i.e., in their preference for which format, or bundle of formats, they like to shop in.Our research question, and model, is similar in spirit to prior research in marketing on how consumers choose assortments. Despite this similarity, our research makes some substantive and methodological contributions to the literature. Substantively, we examine the issue of the choice of channel assortments by consumers across a variety of product categories. We believe this is an important question and one that has not been examined earlier. From a methodological point of view, our model adds to earlier work by specifying the utility of an assortment as a sum of the deterministic and stochastic components of the utilities of its members. This contrasts with previous research where only the deterministic components of the utilities of the component brands of an assortment are added and the relationships between their random components are not accounted for.We calibrate the model on data regarding the format choices of households. In order to control for potentially similar format preferences across purchases of different categories we specify the model to allow for correlation between format preferences over the choice history of each household. Our results suggest that there are four segments of consumers that differ in their preference for different types of formats.     

Brand inequalities in the loyalty-effects of purchases

Choice models in marketing have generally included the effects of loyalty on individual choice behavior.Loyalty has been typically measured as proportion of purchases or as an exponentially smoothed index of past purchases. An underlying assumption of both measures is that all brands gain the same increase in loyalty with a purchase. However, such an assumption may not hold when the competing brands are not comparable. We propose a new exponential smoothing measure which incorporates brand-specific parameters for the loyalty effects of purchases. A choice model, calibrated on individual level data for the detergent category, is used to compare the proposed measure with its traditional version. This comparison reveals that the new approach improves both the fit and predictive performance of the choice model. The results also suggest that the loyalty effects of purchases are likely to be lower for brands which are purchased on price promotions and higher for expensive brands.     

Loyalty differences in the use of internal and external reference prices

Recent findings in reference price research suggest that consumer characteristics may affect whether they use an internal reference price (IPR) or an external reference price (ERP) in price judgments. In this paper, we investigate the role of one such characteristic, brand loyalty, in the use of either type of reference price. Specifically, we employ a latent class-type approach to divide consumers on the basis of their brand loyalty into an ERP and an IRP segment. Analysis of the margarine and liquid detergents categories shows that consumers who are highly loyal to a brand are likely to use external reference prices whereas less brand-loyal consumers rely on internal reference prices. We discuss the implications of this finding and suggest directions for future research.     

Relative importance of online versus offline information for Internet purchases: Product category and Internet experience effects

Across three studies we examine the relative importance of online versus offline information for Internet purchases. Study 1 reveals that the relative importance of online information is higher for utilitarian products (computer hardware and software) than for hedonic products (books, music, and movies). Study 2 shows that, in the case of online purchases, the relative importance of online information decreases with increasing consumer Internet experience. Consequently, offline information becomes relatively more important for consumers with high levels of Internet experience. In addition, the relative importance of online information is higher for utilitarian products than for hedonic products, supporting Study 1 results. Study 3 suggests a possible mechanism for the effect of Internet experience on decreasing importance of online information, showing that consumers' trust of online search engine information decreases with increasing Internet experience. We conclude with implications of our results for firms that sell products on the Internet.     

The palette that stands out: Color compositions of online curated visual UGC that attracts higher consumer interaction

Photos posted by consumers on social media, like Instagram, often include brands. Despite the substantial increase in such photos, there have been few investigations into how prospective consumers respond to this visual UGC. We begin to address this gap by investigating the role of the color compositions of visual UGC in consumer response. Consumer response is operationalized as the click-rate for a photo by a consumer when it is curated on the online site of the brand that it includes. This is the proportion of visitors who click on it for an enlarged view. Composition is operationalized as the specific combination of levels of the photo’s color attributes: hue, chroma, and brightness. Our goal is to identify the color compositions of photos, ceteris paribus, which get more clicks when they are curated. Data for our investigation comes from clicks over a one-year period on photos posted on Instagram curated by fifteen brands in six product categories on their sites. We assume Beta distributed proportions and calibrate a Beta regression using MCMC methods for our investigation.We find that click-rates are higher for photos that include higher proportions of green and lower proportions of red and cyan. We also find that chroma of red and blue are higher in photos with higher click-rates. Findings from our research led the sponsoring firm to modify its proprietary curation algorithm for client brands. The firm informed us that, post-modification, there has been a substantial increase in click-rates of curated photos for brands in several categories.     

Google or BizRate? How search engines and comparison sites affect unplanned choices of online retailers

Recent trends in online retailing suggest that “33% of buyers often or sometimes make unplanned purchases” [Annual Retail Consumer Survey Report. Jupiter Research 2003.]. Findings based on online retailing trends also suggest that both search engines and infomediaries are beginning to play a strong role in leading consumers to online retail sites. Additionally, some practitioner studies find that about half of online consumers use comparison shopping sites before choosing a retailer. Retailers therefore need to understand whether search engines or infomediaries play a stronger role in bringing unplanned consumers. This is the issue that we investigate in this research. Our results indicate that, retailer and consumer factors, category characteristics and the consumer's past relationship with the retailer do play a role in the store choice decisions of online consumers. Between search engines and infomediaries, however, search engines play a far stronger role than infomediaries. The effect is, in fact, more than twice that of infomediaries.     

Accounting for heterogeneity and dynamics in the loyalty-price sensitivity relationship

Marketers have been interested in the relationship between brand loyalty and price sensitivity for many years and have examined whether loyalty reduces consumer price sensitivity. The results, to date, indicate that loyalty does indeed raise the price that consumers are willing to pay for a brand. Other than this broad finding, however, there has been little research in the literature regarding whether and how this relationship varies across consumers and product categories and, within consumers, over time. This is the issue that we investigate in this paper. Specifically, we examine whether the price sensitivity-loyalty relationship is heterogeneous and dynamic. We propose an approach wherein the price sensitivity parameter of a brand choice model is specified as a function of loyalty with three parameters. The first parameter of this function represents the maximum possible reduction in price sensitivity due to loyalty. The second parameter affects the type and shape of the relationship between price sensitivity and loyalty. In particular, depending on the value of this parameter, the relationship could be non-existent, follow a concave shape, indicating decreasing response to increases in loyalty, or be S-shaped to capture the case of increasing response initially followed by decreasing response subsequently. Finally, the third parameter captures the rate at which price sensitivity falls as loyalty increases.We use the proposed approach to investigate the relationship in four frequently purchased categories. In each category, we select a sample of households and calibrate the model on the choices of all the households in the sample. We next employ an Empirical Bayes approach to obtain household-level estimates of all the parameters. These parameters are then used to assign each household in each category to a no response or concave or S-shaped response groups. Within each of these three groups, we assign each household to one of four different response level and rate segments, that is, high response-high rate, high response-low rate, low response-high rate, and low response-low rate. Each of these segments differs in the response level, that is, the maximum reduction in price sensitivity as loyalty reaches a maximum—and the response rate, that is, how quickly price sensitivity falls with increases in loyalty.Following the assignment of each household to a segment in each category, we pool the households across all four categories and calibrate a membership function. This function explains households’ membership in different segments in terms of product category characteristics, household demographics, the household’s responses to price, display, and feature promotions and the evolution of loyalty of the household.Our findings suggest that the nature of the loyalty-price sensitivity relationship does vary across consumers as well as over time. Specifically, the concave response is more likely than the S-shaped response or the absence of a response. We find that the S-shaped response is not related to responsiveness to in-store promotions. It is, however, associated with a slower growth in loyalty to a brand as it is purchased. The concave response, on the other hand, is associated with responsiveness to feature promotions but is unrelated to how loyalty to brands evolves with their purchases.We also find that demographic characteristics are related to the behavior of the concave and S-shaped responses. Specifically, for the S-shaped response, household demographics are related to both the maximum level of the curve as well as its rate of growth. In particular, the curve grows faster with age and its maximum increases with the weekly working hours of the household. In the case of the concave response, high income and more working hours raise the maximum level that the curve achieves. Its rate of growth, however, is unaffected by demographics.We also provide several managerial implications for loyalty and promotional programs based on our findings. Specifically, our first finding—that the loyalty-price sensitivity relationship is dynamic—suggests that, rather than having promotional programs, where the value of the price promotion is fixed and some consumers are targeted with the promotion while others are not, managers should have an entire schedule of price promotions with each level of promotions targeting consumers at a different loyalty level.Our second finding that the nature of the loyalty-price sensitivity relationship is heterogeneous across consumers suggests that designing loyalty programs on the basis of crude classifications such as loyals and non-loyals is not appropriate. Instead, households that are identified as loyal, need to be further divided based on how the loyalty affects their price sensitivity. Promotional programs should then be based on the specific type of relationship that a household exhibits.The third finding that the reductions in price sensitivity to loyalty can exhibit an S-shaped or a concave pattern also has an interesting managerial implication. Specifically, given the differences between the two patterns in how long it might take a consumer to reach a point where s(he) is willing to purchase a brand due to loyalty rather than due to a price promotion, and hence be a profitable customer, it may be preferable for managers to invest more in consumers who exhibit a concave rather than an S-shaped response.Finally, our result that different categories may exhibit different patterns of the relationship between price sensitivity and loyalty implies that each category needs to be analyzed by itself for what the nature of the loyalty-price sensitivity relationship is likely to be so that the most appropriate program for that category can be developed.