Comparing automated text classification methods

Online social media drive the growth of unstructured text data. Many marketing applications require structuring this data at scales non-accessible to human coding, e.g., to detect communication shifts in sentiment or other researcher-defined content categories. Several methods have been proposed to automatically classify unstructured text. This paper compares the performance of ten such approaches (five lexicon-based, five machine learning algorithms) across 41 social media datasets covering major social media platforms, various sample sizes, and languages. So far, marketing research relies predominantly on support vector machines (SVM) and Linguistic Inquiry and Word Count (LIWC). Across all tasks we study, either random forest (RF) or naive Bayes (NB) performs best in terms of correctly uncovering human intuition. In particular, RF exhibits consistently high performance for three-class sentiment, NB for small samples sizes. SVM never outperform the remaining methods. All lexicon-based approaches, LIWC in particular, perform poorly compared with machine learning. In some applications, accuracies only slightly exceed chance. Since additional considerations of text classification choice are also in favor of NB and RF, our results suggest that marketing research can benefit from considering these alternatives.     

Shut the fridge door! HRM alignment,job redesign and energy performance

Anchored within the strategic HRM and alignment literature, and drawing on efficiency and legitimacy perspectives of organisational behaviour, we investigated a HRM intervention targeted at energy reduction goals in a large multinational retailer. The HRM intervention was focused on embedding the environmental and economic performance goals of the firm within the workplace through redesigning the job so that energy tasks were aligned with training and performance management systems, as well as organisational performance goals. Using a randomised control trial design, we tracked changes in energy behaviours and energy consumption in 769 retail stores (685 in the intervention condition, 84 in the control condition). The findings provide evidence that changing the alignment of HRM practices can influence both worker behaviour and organisational outcomes, including environmental outcomes. This work contributes to debates concerning the impact of HRM alignment on both the work and organisational performance context.     

Durability Versus Concentration as an Explanation for Price Inflexibility

We document the extent of price rigidity across United States manufacturing industries in the 1980s and early 1990s and compare rigidity across different phases of the business cycle. We measure price rigidity in three ways – each under four different sets of assumptions. We take an approach that relies on disaggregated data; we look at price patterns for over 4000 individual manufactured commodities. Both durability and seller concentration are found to be important factors explaining differences in price rigidity across industrial product classes. Using our data, we replicate the regression results found in Carlton (1986) that were based on actual transaction prices from the 1960s.     

A Spatial Autocorrelation Approach for Examining the Effects of Urban Greenspace on Residential Property Values

This paper presents spatially explicit analyses of the greenspace contribution to residential property values in a hedonic model. The paper utilizes data from the housing market near downtown Los Angeles. We first used a standard hedonic model to estimate greenspace effects. Because the residuals were spatially autocorrelated, we implemented a spatial lag model as indicated by specification tests. Our results show that neighborhood greenspace at the immediate vicinity of houses has a significant impact on house prices even after controlling for spatial autocorrelation. The different estimation results from non-spatial and spatial models provide useful bounds for the greenspace effect. Greening of inner city areas may provide a valuable policy instrument for elevating depressed housing markets in those areas.     

The economic function of credit rating agencies – What does the watchlist tell us?

Credit rating agencies do not only disclose simple ratings but announce watchlists (rating reviews) and outlooks as well. This paper analyzes the economic function underlying the review procedure. Using Moody’s rating data between 1982 and 2004, we find that for borrowers of high creditworthiness, rating agencies employ watchlists primarily in order to improve the delivery of information. For low-quality borrowers, in contrast, the review procedure seems to have developed into an implicit contract á la Boot et al. (2006), inducing the companies “on watch” to abstain from risk-augmenting actions. The agencies’ economic role hence appears to have been enhanced from a pure information certification towards an active monitoring function.     

Union Organizing as a Mobilizing Strategy: The Impact of Social Identity and Transformational Leadership on the Collectivism of Union Members

This article investigates the effect of union organizing as a mobilizing strategy on the collectivism of union members. We examine the impact of a worker's social identification with fellow members and the transformational leadership qualities of the local union representative. We employ regression analysis with tests of mediation to analyse the survey responses of c. 1,000 rank and file members of a major professional union, collected in July 2004 during a mobilization campaign. Social identification and transformational leadership were associated with members' union loyalty and willingness to work for the union. Social identification acted as a mediating variable in both cases.     

Intolerable nuisances: some laboratory evidence on survivor curve shapes

The fraction of a user population willing to tolerate nuisances of size x is summarized in the survivor curve S(x); its shape is crucial in economic decisions such as pricing and advertising. We report a laboratory experiment that, for the first time, estimates the shape of survivor curves in several different settings. Laboratory subjects engage in a series of six desirable activities, e.g., playing a video game, viewing a chosen video clip, or earning money by answering questions. For each activity and each subject we introduce a chosen level \(x \in [x_{\min }, x_{\max }]\) of a particular nuisance, and the subject chooses whether to tolerate the nuisance or to switch to a bland activity for the remaining time. New non-parametric techniques provide bounds on the empirical survivor curves for each activity. Parametric fits of the classic Weibull distribution provide estimates of the survivor curves’ shapes. The fitted shape parameter depends on the activity and nuisance, but overall the estimated survivor curves tend to be log-convex. An implication, given the model of Aperjis and Huberman (SSRN, doi:  10.2139/ssrn.1672820, 2011), is that introducing nuisances all at once will generally be more profitable than introducing them gradually.