Purpose -This paper aims to examine which behaviour or set of behaviours customers are prone to follow in double deviation scenarios (i.e. consumption experiences in which customers face both the initial service failure and a failed service recovery), as well as how customers' perceptions of the problem and the firm's recovery efforts may influence these behaviours. Design/methodology/approach -The paper uses multinomial logit models with random coefficients to test the proposed model. Findings -Magnitude of service failure, explanations, apologies, perceived justice, anger and frustration felt by the customer, and satisfaction with the service recovery have significant and different effects on customers' choice of a type of response. Research limitations/implications -Additional research should try to determine the effects of different variables and their potential interactions. Further work incorporating different subjects, service settings or additional combinations of complaining behaviours is needed to validate the results of this investigation. Practical implications -This study highlights the importance of effective management of consumer responses to double deviations. Even when it is not possible to respond to customer complaints the first time, firms can learn from double deviations. Furthermore, new market entrants and competitors who want to capture consumer switchers should recognise what happened and try to avoid making the same mistakes. Originality/value -This study is the first to examine the consequences of double deviations by considering the multi-dimensional nature of complaint behaviour and the existence of simultaneous responses. This study is based on analyses of real service failures and recovery strategies and actual customer behaviour.

This study quantifies for the first time the influence of flow regulation on the river thermal behaviour of an ungauged basin located in central-eastern Argentina. A 30-day data set of continuous summer hourly data was assembled for eight water temperature gauging sites deployed along the main channel upstream and downstream from the impoundment. Analysis methods include descriptive statistics of daily temperature data, classification of diurnal regimes by relative differences in the 'shape' and the 'magnitude' of the thermographs (RSMC), and quantification of the climatic sensitivity of water temperature regimes using a sensitivity index. Results revealed that temporal fluctuations in water temperatures were linked to meteorological drivers; however, spatial variability in the shape and the magnitude of the thermographs revealed the effects of the dam in regulating river thermal behaviour downstream. Water temperatures immediately below the dam were reduced notably; diurnal cycles were reduced in magnitude, delayed in timing, and revealed overall climatic insensitivity and high temporal stability in regime shape. Dam effects persisted along the 15-km stretch monitored, although declined in the downstream direction. These findings provide new scientific understanding about the river water quality and inform river management about potential shifts in summer water temperature with great implications for the diversity and lifecycles of Neotropical river fauna. The use of the RSMC and sensitivity index approaches in water temperature assessment is novel and has wider applicability for quantifying river thermal regimes and their sensitivity to drivers of change over a range of temporal and spatial scales. Figure 4. Longitudinal profiles of mean, maximum, and minimum temperatures recorded over the 30-day monitoring period. Numbers indicate stationto-station thermal gradients ( C km À1 ).Figure 7. Distribution of composite shape and magnitude classes over curves of daily mean temperature by site (right) and spatial distribution of the dominant composite regime classes of air and water temperature (left).530 A. CASADO et al.