Resource Similarity,Market Commonality,and Spatial Distribution of Hotel Competitive Sets

Abstract

Index scores and competitive sets (compsets) play a critical role in the performance and evaluation of hotels. The reliance on these metrics has drawn skepticism in recent years as competitive sets may be opportunistically chosen, creating bias in performance evaluation. Drawing from the principal–agent theory and the theory of incentives, we explore whether the distance of the competitors chosen for a hotel’s compset influences revenue per available room (RevPAR) index scores. Based on the concepts of resource similarity and market commonality, we develop a novel mathematical model through which we empirically analyze a large dataset of 10,000 compsets. We find evidence that competitor distance influences index performance and that this relationship is bidirectional. Results show that hotels that outperform the competition may use distance to inflate RevPAR indices, while those that underperform may use distance to further reduce scores. These conflicting results may be reflected from the reverse motivations of the stakeholders.

Keywords

compsets hotels RevPAR index opportunism bias principal–agent

Introduction

Benchmarking against the competition is a key element in the evaluation of hotel performance (e.g., Buckhiester, 2011; Smith & Zheng, 2011). These comparisons support stakeholders’ short-term operational decisions in activity domains such as marketing and revenue management and are also central for the longer-term strategic decisions of investment, asset management, brand positioning, and product mix (Love et al., 2012; Morgan & Dev, 1994). Specifically, a large and growing number of hotels in the United States, and globally, compare hotel key performance indicators (KPIs) such as the average daily rate (ADR), occupancy, and revenue per available room (RevPAR) with the averages of their competitive sets (compsets) (Chipkin, 2007; Cross et al., 2009; Enz et al., 2014; Kim & Canina, 2011; Schwartz et al., 2017; Watkins, 2017).

However, several publications have indicated that the hotel industry should be more vigilant with its reliance on compsets (Enz et al., 2001; Kim & Canina, 2011; Slattery, 2002). For example, Webb and Schwartz (2017) demonstrated how the decision to include, or exclude, a subject hotel in a compset could bias its RevPAR index (a hotel’s RevPAR relative to an aggregated grouping of its compset), proved that high levels of uncertainty call for opportunistic exclusion tactics, and conjectured about the negative impact that the use of these relative performance indices could have on the hotel’s inclination to collaborate with competitors, despite the recognized advantage that such a co-opetition might have.

The current study continues this line of research, looking critically at the lodging industry’s compset practices by introducing the well-established performance measures’ concepts of market commonality (MC) and resource similarity (RS) in the context of hotel compsets’ bias. More specifically, the study explores the issues of the distance-related opportunistic selection of the compset hotels, that is, their distance from the subject hotel, and develops the analytical model to explain how these concepts are related. Based on the analytical formulation, and given insights from the principal–agent model and the theory of incentives (e.g., Laffont & Martimort, 2009; Stiglitz, 1988), the study conjectures that by adding distant hotels to a compset, one could, more effectively, affect the outcome of the comparative performance measure. Stakeholders, aiming to make the hotel “look better” by selecting lesser performing hotels for their compset or, conversely, make the hotels look worse by selecting better performing hotels are likely to include more distant hotels. As we demonstrate in the model, developed later in the article, the farther the decision makers search for compset hotels, the more likely they are to find the hotels they are looking for, that is, better (or worse) performing hotels compared with the subject hotel. The analytical formulation is followed by an empirical investigation of a large data set of 10,000 hotels and their primary compsets. The findings suggest that some hotels might indeed use distance to affect their RevPAR index. That is, distance appears to shift RevPAR index scores in potentially desirable directions, that is, away from 100.

The theoretical contribution of this study is threefold.

The study is first to discuss, and formally (mathematically) model, the connection between the concepts of MC and RS to the hotel compsets’ vulnerability, and to opportunistic behavior in the framework of principal–agent theory. Despite being recognized in the general literature as the building blocks of strategic competitive analysis, these two concepts of MC and RS were not yet incorporated in the discussion on bias in hotel compsets.

It is first to show the connection between spatial distributions of compset membership and the potential to purposely influence the outcome, that is, to bias the performance indices.

It is also first to suggest, and empirically test, the notion that with the industry’s complex control structure of multiple stakeholders, that is, owners, management company, and brand, contradicting interests might be present and at work. As such, and depending on their power to control the compset’s membership, these various players might influence opportunistic compsets in opposing directions, that is, either increase or decrease the subject hotel’s performance indices.

For industry practitioners, this study offers an insight on yet another area of compset vulnerability—the inclusion of distant hotels in the compset. As such, it provides an opportunity to engage in a discussion on what practical steps could be taken, either at the hotel property/corporate levels or with the performance reports providers such as STR, to reduce the damage caused by the opportunistic behavior, and consequently increase the indices validity for decision makers.

The following section develops the model, connecting the concepts of MC and RS to the makeup of compsets, and to the compset members’ distance from the subject hotel.

Literature Review

Desirable Versus Opportunistic Makeup of a Competitive Set

The strategic management literature (e.g., Peteraf & Bergen, 2003), as well as the lodging industry’s practices (e.g., Otainsight, 2018; Watkins, 2015), suggest that an adequate compset should be based on both RS and MC. Broadly speaking, RS refers to characteristics such as hotel capacity, location, the age of the property, class, and brand affiliation (e.g., Kim & Canina, 2011). Members of the subject hotel’s compset are expected to have similar levels of these characteristics. MC (Chen, 1996) is defined from the hotel customers’ perspective in that the compset hotels serve customer segments with similar needs and wants. Those needs and wants include the hotel’s location, and it implies that some customers include both the subject hotel and compset hotels in their consideration set for a specific future hotel stay. In other words, it implies that the subject hotel and the members of its compset directly compete with each other over a substantial number of “shared” potential customers.

This study proposes that the difference between the RS and MC traits affects the distribution of potential adequate compset members around the subject hotel. In other words, the number of appropriate candidates for inclusion in the subject hotel’s compset, which could be found within a certain distance, depends, to some extent, on how one defines “appropriate”: RS, MC, or both. We begin by defining H_All, the total number of hotels around the subject hotel, to be

H_{A l l} = α π r^{2}

(1)

where r denotes the radius of a circle around the subject hotel, $π r^{2}$ is the space of that circle, and α (>0) is a coefficient denoting the “density” of hotels within that circle around the subject hotel.

Now, consider a hotel that is only concerned with RS. For the sake of simplicity, one could assume that resource similar hotels (RSH) are evenly distributed in the space around the hotel.¹ If β (>0) is the portion of RSH within that group of hotels, then H_rs, the number of hotels that are adequate for inclusion in its compset, is,

H_{r s} = α β π r^{2}

(2)

If RS is the sole criteria (that is, the hotel ignores the need for MC), the further it searches, the larger the pool of adequate candidates for compset membership. Note that in this case of RSH only, the growth of the pool is positively monotonic: For every additional mile (of radius), the number of adequate RSH increases by

\frac{\partial H_{r s}}{\partial r} = α β π 2 r

(3)

An ideal compset should, as discussed above, consider both RS and MC. Given the nature of the hotel industry product, we argue that MC is also location dependent. It is well established that hotel guests select hotels based, in large part, on the hotel’s location (Baruca & Čivre, 2012; Fox, 2011; Rivers et al., 1991; Tsaur & Tzeng 1996; Wong & Chi-Yung, 2002). That is, to be considered, and ultimately chosen, by customers, the hotel needs to be in proximity to the customer’s destination and/or intended activity. Hence, from the MC perspective, location is related to distance and less to characteristics such as view. This implies that the more distant hotels are from each other, the less likely they are to “share” potential customers because their location caters to a different set of people. Hence, MC decreases with distance and therefore can be described using a decaying function such as e^−Δr. In the context of hotel compsets, a decaying function reflects the notion that as the radius around the subject hotel increases, the number of MC hotels, added to the pool of adequate hotels, is increasingly smaller and reaches zero at some distance. If an ideal compset hotel should have both RS and MC, it follows that the number of adequate hotels within a radius around the subject hotel is given by

H_{m c & r s} = α β π r^{2} e^{- r θ}

(4)

where $θ$ is a coefficient denoting the rate of “decay,” that is, the decrease in the number of MCH as the distance from the subject hotel increases. Specifically, for every additional mile (of radius), the number of adequate hotels in this compset, where the hotels are both RS and have MC, is given by

\frac{\partial H_{m c & r s}}{\partial r} = - α β π r (θ r - 2) e^{- r θ}

(5)

Figure 1 demonstrates these relations between the number of potential compset hotels and the radius of the considered area around the subject hotel. The concave form of the lower line in Figure 1, depicting the “ideal compset” pool, demonstrates this relation: As the considered range around the subject hotel increases, fewer hotels are added to the pool of compset appropriate hotels. From a practical point of view, it means that for an “honest” hotel there is little incentive to keep increasing the distance, because fewer hotels are added, the numbers drop at an increasing rate, and at some distance none at all are added to the pool of proper hotels to consider.

Figure 1

Potential Pool of Compset Hotels as the Distance Increases

It is important to note that in the hotel context, location can be used for both RS and MC purposes, and leveraged simultaneously or independently. For instance, MC may dictate that competitors are based solely on the location of a destination as mentioned above. In this case, hotels in New York and San Francisco are unlikely to be competing for the same customers at the same time. However, from an RS perspective, location may also be used for property-based attributes such as distance from the beach or an iconic location. In either instance, the case can be made for the importance of location in compset formulation, using RS and MC characteristics, independently or simultaneously. Recognizing the importance of distance and location to compset membership, the focus of this study is to understand the potential impact that distances may have on performance metrics and the potential issues associated with distant compsets.

Compset Distance and Opportunism

Hotels’ relative KPIs such as the RevPAR Index are important factors in determining management compensation. As stated by Mainzer (2004):

The hospitality community has encouraged the development of competitive indices to benchmark performance across the industry. The most widely used benchmarking index is revenue per available room (RevPAR) . . . The RevPAR measure has become the standard for evaluating hotel revenue performance and is even used to determine compensation levels for hotel directors of sales and general managers.

Management’s financial bonuses are one mechanism used to ensure that agents (management) are making decisions that are in the interest of the principals, that is, the owners (e.g., Garen, 1994; Murphy, 1999; Smirnova & Zavertiaeva, 2017). At the same time, numerous studies show that agents often act opportunistically in a manner that will increase their bonuses, often at the expense of the owners’ interest. For a comprehensive review of the principal–agent framework in the context of compset opportunistic composition, see Webb and Schwartz (2017). It follows that the person, or a unit, selecting the compset’s hotels could leverage the ability to choose distant hotels in order to include hotels in a way that will tilt the outcome in a desirable direction. Consider, for example, a subject hotel that wishes to have a compset, which is less likely to perform as well, so that its relative KPI (e.g., RevPAR Index) is consistently above the compset’s average. This subject hotel might find it easier to identify these “make me look better” hotels if it searches farther away. As the considered radius increases, the pool of “really adequate” hotels quickly reaches a plateau and remains relatively small. This is demonstrated by the lower curve in Figure 1 (that ideal pool for compset hotels). However, the situation is very different if the subject hotel disregards the MC requirement. Consider $γ$ ( $0 \leq γ \leq 1$ ) to be the portion of hotels (in that pool of RSH) with RevPAR lower than that of the subject. The number of “make-me-look-better” hotels is then given by

H_{r s c} = α β γ π r^{2}

(6)

and, in this case of no requirement for shared customers, for every additional mile of radius, the number of adequate hotels in this compset (RSH only) is

\frac{\partial H_{r s c}}{\partial r} = α β γ π 2 r

(7)

The equations, as well the corresponding lines in Figure 1, demonstrate the following: When the subject hotel disregards MC, it is able to find an increasingly larger number of hotels with similar resources because it considers a larger area around its location. The larger that area, the more number of hotels to consider for the compset, and that number of hotels grows exponentially with the radius. More important, the number of hotels that perform worse than the subject hotel also grows exponentially with the radius, making it easier for an opportunistic hotel to add “desirable” hotels to its compset, that is, hotels that are resource similar but subperform compared with the subject hotel.

The same logic holds when the goal of the opportunistic behavior is flipped—that is, when the goal is to make the subject hotel look less successful. This type of reversed motivation could exist when the stakeholders that decide on which hotels to include in the compset are on the other side of the contract, that is, on the bonus-paying side. Hotel owners, or perhaps the management company, might prefer a consistently lower RevPAR index if the bonuses they pay to the property-level executives are tied to that index. Note that the analysis above applies to this reverse motivation case of opportunistic behavior, and the only difference is that we replace $γ$ with ( $1 - γ$ ) in Equations 6 and 7 above.

To summarize, opportunistic stakeholders who aim to look better by selecting lesser performing hotels for their compset, or make the hotels look worse by selecting better performing hotels, are likely to include more distant hotels. This is likely to happen because the farther they go to select hotels for the compset, the more likely they are to find the hotels they are looking for—either better or worse performing, but resource similar. However, if opportunistic intentions are not the motivating factor for including more distant hotels, one would expect to find no correlation between the subject hotel’s RevPAR index and the average distance from their compset. Accordingly, this study hypothesizes the following:

Hypothesis_1o: Average competitor distance does not have an effect on the variability of a hotel’s RevPAR index.

Hypothesis_1a: Average competitor distance does have an effect on the variability of a hotel’s RevPAR index.

Note that the use of an average competitor distance represents a conservative/cautious approach. Consider the following: Given the small size of typical compsets, including one or two distant hotels by the interested agent is sufficient to tilt the index in a desirable direction. In this regard, analyzing the relations using the longest distance in the compset (instead of the average) is bound to generate considerably stronger results. Hence, demonstrating the relations between the average compset distance and RevPAR index is a strong indication that the findings are robust. The hypothesis above is about the expected increase in opportunities to select hotels with desirable level of performance. The next two hypotheses are about indications of opportunistic behavior. We refer to two situations: hotels with RevPAR index above 100 and hotels with RevPAR index below 100. For the successful hotels with an index above 100, we hypothesize the following:

Hypothesis_2o: The average competitor distance has no effect on RevPAR index scores.

Hypothesis_2a: The average competitor distance has a positive effect on RevPAR index scores.

For hotels with RevPAR index lower than 100, we expect the opposite:

Hypothesis_3o: The average competitor distance has no effect on RevPAR index scores.

Hypothesis_3a: The average competitor distance has a negative effect on RevPAR index scores.

In the following section, we test these hypotheses using a large data set of U.S. hotels’ primary compsets.

Empirical Analysis

Methodology

The study analyzes a data set provided by STR, a lodging industry global leader in collecting performance data and disseminating relative performance reports. The initial (anonymized) data contained approximately 10,000 hotels, with information about location, occupancy, ADR, and RevPAR scores of the compset, as well as the distances of the compset hotels from the subject hotel. This random sample of hotels is limited to hotels in the United States. It was stratified to represent the STR client base of hotels in terms of location, class, and size. The average competitor distance was 8.33 miles, while 60% of properties had an average distance under 3 miles. To mitigate the impact of atypical compsets with exorbitant distances, compsets with average competitor distance greater than 50 miles were removed from the analysis. In total, 187 properties, or less than 2% of the data, were removed. The final sample consisted of 9,706 properties with the distribution of the sample characteristics provided in Table 1.

Table 1

STR Property Characteristics

STR Property Characteristics	Properties	Percentage
Operation
Chain management	1,344	14
Franchise	7,834	81
Independent	531	5
Class
Luxury	241	2
Upper upscale	640	7
Upscale	1,412	15
Upper midscale	2,932	30
Midscale	1,776	18
Economy	2,708	28
Location
Airport	605	6
Interstate	1,472	15
Resort	558	6
Small metro/town	1,956	20
Suburban	4,082	42
Urban	1,036	11
Rooms
<75	3,407	35
75-149	4,673	48
150-299	1,200	12
300-500	309	3
>500	120	1

The analysis focuses on two questions. First, we check if the variability of the RevPAR index (around the median) is associated with the compset’s distance. This is accomplished by comparing the average RevPAR indices of hotels based on the distance of the compset hotels from the subject hotel. In other words, we explore whether subject hotels within groups of different distances have different levels of RevPAR index variability. The motivation for this analysis is that a larger variability of the RevPAR index means that for the opportunistic hotel, there are more options for compset composition and, as such, the ability to behave opportunistically is enhanced. Specifically, we group the hotels into three compset distance groups: hotels with a compset where the average distance of the compset hotels from the subject hotels is less than 3 miles, distance of 3 to 10 miles, and more than 10 miles. We then calculate the RevPAR index of the upper quartile, the lower quartile, as well as the interquartile. The H spread (interquartile) is a common measure of variability and is the difference between the first and third quartile. To test the significance of the variability among the distance groups, the Wilcoxon rank test was implemented. Specifically, the samples were split above and below the median RevPAR index to statistically test the differences in variability. This allows for interpretation as to whether index variability exists among different distance groups, as well as if it is unidirectional or bidirectional. Formally,

H₀: The distribution of RevPAR indexes is equal across distance groups.

H₁: The distribution of RevPAR indexes is not equal across distance groups.

While the first question is about the availability of opportunities, and its relation to compset distance, the second question explores whether there are indications of actual opportunistic behavior—that is, “Is there empirical evidence that the opportunities presented by the alleged increased variability of the RevPAR index are put to use by hotels?” To answer this question, we first categorize the hotels as A (above) or B (below) type. An A hotel performs better than its compset, that is, it has a RevPAR index above 100. A B hotel performs below its compset average, that is, it has a RevPAR index below 100. Within each of these two groups, we select two subsets of 500 hotels each—the best and worst of each. That is, the group of 2,000 hotels, analyzed in this part of the study, is composed of the 500 hotels with the highest RevPAR index among the A hotels, the 500 hotels with the lowest RevPAR index among the A hotels, and the same two types of groups among the B hotels. We then compare the average distance of the compset hotels from the subject hotel across these four groups of 500 hotels using a t test to identify statistically significant differences in compset distance among the two categories of A and B.

While the initial analyses provide a general overview of the influence distance may have on RevPAR index scores, these analyses do not control for the various factors that may influence initial compset selection. Two ordinary least squares regression models were used to measure the influence of distance on RevPAR index scores. In both models, the RevPAR index was used as the dependent variable, while the average distance of the properties included in the subject hotels’ compset was used as the predictor of interest. To differentiate the directional effects of distance on compset scores between A and B hotels, an interaction term between compset distance and Group A was introduced to account for differences in slope and direction between the two groups. To ensure that the distance relationships were robust across the various property types, the operation (franchise as reference), the hotel class (economy as reference), the location (small/metro as reference), and the size of the property were introduced as control variables in the second model.

Results

We find (Table 2) that the RevPAR index variability around the median, in the continuum of RevPAR indices, increases with the distance of the compsets’ hotels from the subject hotel location. The farther apart the compsets’ distance, the larger the interquartile, or other interpercentile levels for that matter. It means that, on average, the RevPAR index of hotels with closer compsets is closer to 100—the theoretical fair share of each hotel within the market.

Table 2

First and Fourth Quartiles, and the Interquartile Spread of the RevPAR Index

	Average RevPAR Index
Distance in miles:	0-3	3-10	10-100
Top 25% hotels (highest RevPAR index in the sample)	115	119	121
Bottom 25% hotels (lowest RevPAR index in the sample)	81	79	73
Interquartile spread	34	40	48
N: No. of hotels in the sample	5,983	2,699	1,024

As clearly shown in Figure 2, the size of the box plots’ interquartile range increases with distance. The results of the Wilcoxon tests for difference in RevPAR distribution among the distance groups show that distance does affect the distribution of RevPAR scores. The sample testing scores above the median shows that both the 3- to 10- and 10- to 50-mile ranges have a significantly higher distribution of RevPAR scores than those in the 0- to 3-mile range. For the index scores below the median, significantly lower RevPAR distributions are found in compsets with an average competitor distance greater than 10 miles. These findings of higher distribution of RevPAR index in more distant compsets support the theoretical notion and derived hypotheses as suggested in the first part of this study. The distance appears to provide an opportunity to nudge the RevPAR index in a certain direction—the larger the area the higher the variety of RevPAR scores to choose from, and thus the easier it is to influence an index in a desirable direction.

Figure 2

RevPAR (Revenue per Available Room) H Spread and the Compset Distance

As discussed in the Methodology section above, we divide the sample into two groups—hotels with RevPAR above 100 and below 100—and analyze the two extreme groups of the 500 highest and lowest RevPAR index hotels within each of these two categories. We find that the direction of the difference in performance (RevPAR index) across compset distances flips between the two groups (A and B). This is demonstrated by the slope of the lines in Figure 3 where the relation is convex (V shaped), and where the bottom of the V is near the 100 RevPAR index mark. The A group (i.e., hotels with RevPAR index above 100) appears to benefit from distant compsets, while the B group (hotels with RevPAR index below the average of their compsets) appears to have lower RevPAR index the farther away their compset.

Figure 3

Distance of Compset Hotels Among Four Groups of 500 Hotels Each, Categorized by Their RevPAR (Revenue per Available Room) Index

The top 500 performing hotels (i.e., hotels with the highest RevPAR index) among hotels that outperform the compset have a statistically significant more distant compset, compared with the bottom 500 performing hotels among hotels that outperform their compset. This relation flips for B type hotels, that is, for hotels that did not outperform their compset. The compset members of the 500 hotels at the bottom of the performance scale among the B hotels are further away, compared with the compset members of the top 500 performing A hotels. In both instances, the t test used a log transformation of the average compset to satisfy the assumption of normality. The difference between the top 500 and bottom 500 (within each group of A and B hotels) are statistically significant at p < .000.

The results of the regression analysis are provided in Table 3. Model 1 shows a significant negative effect of compset distance on RevPAR index scores. In other words, for each additional mile of average competitor distance, the RevPAR index score decreases by 1.83. Hence, for a compset with an average distance of 10 miles, the RevPAR index score is reduced by 18.3 points. The interaction effect is also significant with a parameter estimate of 3.84. Combining this with the overall negative effect for distance implies that for A hotels, the impact of distance on RevPAR index scores is a net positive of 2.01 for each mile. Overall, the results indicate that the average distance of competitors shifts the index scores further from the expected market share baseline of 100, with directional differences based on whether the properties index is above or below 100. To test the robustness of the results, the control variables were entered into Model 2. The model estimates show the same significance and direction for the distance parameters indicating consistent findings across numerous hotel types.

Table 3

Regression Model

Variable	Model 1	Cohen’s F	Model 2	Cohen’s F
Intercept	99.41 (0.33)*		91.53 (0.82)*
Compset average distance	−1.83 (0.05)*	0.12	−1.49 (0.06)*	0.07
Interaction (Group A × average distance)	3.84 (0.07)*	0.30	3.31 (0.07)*	0.24
NumRooms			−0.03 (0.003)*	0.01
Chain_management			7.53 (0.83)*	0.01
Independent			−9.84 (1.28)*	0.01
Luxury			27.41 (1.94)*	0.02
UpperUpscale			27.26 (1.34)*	0.04
Upscale			24.70 (0.89)*	0.08
UpperMidscale			17.81 (0.71)*	0.06
Midscale			4.00 (0.78)*	0.00
Airport			−3.32 (1.22)*	0.00
Interstate			0.13 (0.88)	0.00
Resort			−1.84 (1.29)	0.00
Suburban			−1.98 (0.74)*	0.00
Urban			−4.46 (1.09)*	0.00
Adjusted R²	0.231		0.3195
N	9,706		9,706

Note: Table 3 shows coefficient (standard deviation). variance inflation factors <2.1.

Significant at .01 level.

Traditional research has focused on p values and statistical significance to identify relationships, generally using a 5% threshold for committing a Type 1 error. However, academics have noted that large sample sizes can generate significant p value results when differences are miniscule (Khalilzadeh & Tasci, 2017). In other words, the differences between groups are not practical, however significant due to the sample size. In these instances, researchers should also consider effect size, which depicts the magnitude of the effect. In essence, effect sizes should be greater than zero (trivial difference) with the larger the effect size identifying a stronger relationship (Lakens, 2013). To test effect size in our models, Cohen’s $f^{2}$ was calculated, which measures the local effect size attributed to each variable (Cohen, 1988; Selya et al., 2012). According to Cohen’s guidelines for local effect sizes, $f^{2} \geq 0.02,$ $f^{2} \geq 0.15,$ $f^{2} \geq 0.35$ represent small, medium, and large effect sizes, respectively (Cohen, 1988). The results of our tests show that the average compset distance had a low effect size ( $f^{2} = 0.12$ ), while the interaction term had a medium effect $(f^{2} = 0.30) .$ Considering the parameters represent distance across groups, it appears that distance had a noticeable effect on RevPAR indices and had a larger effect on Group A hotels than Group B—that is, the magnitude of distance is more pronounced when RevPAR indices are greater than 100. The effect sizes were low or had no effect for many of the control variables. This result is not surprising given that the r² estimates were similar between both models.

These findings appear to support this study’s second and third hypotheses and what the literature on principal–agent models, incentive theory, and observed opportunistic behavior all discuss. The results of the models suggest that after controlling for a number of variables that may influence compset formulation, distance appears to significantly affect RevPAR indexes. Interestingly, the findings also suggest that there are three relevant groups in this framework of analysis and not two (A and B). The extreme group of best among the A hotels is the “true” As. The worst among the B type hotels are the “true” Bs. The two mid-performing groups at the bottom of the V shape graph of Figure 3 (best among the A hotels and worst among the B group) are not that different when it comes to their RevPAR index, and, more important, the difference between the two groups’ average distance of the compsets is not statistically different from zero. This could be interpreted as follows. The 1,000 hotels with RevPAR index close to the median of 100 (either slightly above or slightly below) appear to be the “honest” ones. The distance of their compset is the lowest, and it indicates that that they do not include distant hotels to affect their RevPAR index. With the two remaining groups of extreme values of RevPAR index, it is more likely that distance is used to manipulate the KPI. The “true” A hotels are successful hotels and as such are probably better managing the many aspects of their RevPAR index, from more effective and efficient hotel operations to possibly manipulation of compset membership. Considering distant hotels adds flexibility to their choice, allowing these “shrewd” hotels to include properties that not necessarily perform as well as them, making them look better, and, consequently, increasing the executive bonuses. The findings regarding the “true” B hotels (the 500 hotels with the lowest RevPAR index in the entire sample of 10,000 hotels) are intriguing. A possible explanation is that these hotels might not be good at “managing” in general, and are probably also less mindful, or capable, when it comes to “manipulating” their relative KPIs. Another possible explanation is that these hotels have less control over the decisions about setting up their compset and thus end up with a less favorable compset. As alluded to earlier in the article, sometimes stakeholders such as corporate office, brand, or the owners (Butler, 2016) make that compset membership decision. As stated by Watkins (2015),

Often, hotel ownership is keenly interested in comp set and in providing input to the creation of the set as a way to verify the management team is doing a solid job running the business and bringing the right business to the hotel to make the asset as profitable as possible.

If the bonus is tied up to that compset-based KPI, then that stakeholder has an incentive to “downplay” the KPIs. In that case, the larger flexibility works in reverse, in that it is an opportunity to select from a larger pool of more distant hotels so that the RevPAR index looks worse. More research is needed to fully understand the reasons behind the reverse direction observed with B type hotels.

While the motivation for the formulation of distant compsets was not explicitly tested, the empirical results suggest that distance may be used as one method for a hotel to “look better.” Assuming that bonuses are directly tied to KPI metrics, one may assume that the tenets of agency theory will hold and competitor distance may be used as an opportunity. Regardless of motivation, location (distance wise) is a key element of compset formulation based on the principals of RS and MC. It appears that expanding the compset radius moves index scores in an undesirable direction (further from the baseline of 100). This presents challenges for monitoring performance, and it is recommended that key stakeholders adequately account for distance when formulating their compsets.

Conclusion, Limitations, and Future Research

This study explored the connection between compset distance and opportunistic behavior of subject hotels. The model developed and described is a major contribution of this study. It explores what happens when the commonly accepted requirement of MC between a subject hotel and its compset is lessened. The theoretical analysis shows that this relaxation of the selection rule allows hotels to include more distant hotels in their compset, increasing the flexibility they have, in that it provides a stakeholder with more opportunities to select certain hotels that could affect the subject hotel’s RevPAR index in a desirable direction. We follow up the theoretical model with an empirical investigation, testing the theoretical model’s predictions and formal hypotheses. The analysis of a large data set of compsets suggests that the selection of distant hotels might indeed be associated with opportunistic behavior but that this selective, distance-based, compset construction goes both ways—that is, distant hotels might be selected to increase, or decrease, the relative performance indicator of RevPAR index. We interpret this duality to indicate that the distance is probably used by the two sides of the principal–agent problem. Principals (owners), who have a say in the process, include hotels with higher RevPAR so that the subject hotel relative performance is diminished. This biasing action ensures that the performance-based bonuses they pay to the management are reduced. When management has a say in the process, that is, when they are allowed to affect what hotels are included in the compset, the managers might be advancing their own financial interest by including distant hotels, hotels that are poorly performing, that is, hotels with lower RevPAR, thus increasing the bonuses they receive.

The practical implications for the hotel industry are straight forward. Stakeholders and perhaps the service providing firms such as STR, who have concerns about biased compsets in either direction, should aim to reduce the compset distance. That is, make sure that the compset hotels are located as close as possible to the subject hotel. In some cases, there is a legitimate reason to increase the compset distance. For example, a subject hotel might need to include hotels that are located far away because there are not enough hotels nearby that are resource similar. In this case, it is recommended that the concerned stakeholder carefully scrutinize the distant hotels, and make sure that they have a compelling reason to be included in the compset, not because their RevPAR is such that it nudges the index in a certain direction, a direction that is favorable to the interests of whoever decided to include these distant hotels in the compset. While the principal–agent dynamic is applicable to all stakeholders, the influence of distance can expand beyond bonus incentives to other decisions based on RevPAR index metrics. For instance, revenue management pricing decisions may be calibrated based on feedback from weekly market performance. Similarly, investment decisions such as stock prices and real estate evaluations may be overinflated by including distant competitors. These examples show that biased metrics may have larger implications for the industry and distance must be scrutinized in compset formulation.

This study is a first attempt to explore the question of compset distance and bias in a systematic manner. As such, it has several limitations, and it offers several exciting opportunities for follow-up research. Future studies could explore the circumstantial factors that might be affecting this connection between compset distance and the RevPAR index. Using global data rather than the U.S. based sample used in this study will support a cross-country analysis. Equally interesting is the question of what appears to be reversed, or opposing, motivation to affect the RevPAR index through the inclusion of distant hotels in the compset. We find evidence to support the notion that the distance can be used to either increase or decrease the RevPAR index. Obviously, this is just a possible explanation to the V shape of the relation, and more work, supported by in-depth interviews with stakeholders and decision makers, is required to validate this explanation or identify and support an alternative one. Furthermore, while the theoretical model depicts the influence of distance on compset membership from the MC and RS contexts, future research could explore which component dominates the role of distance in compset formulation. Finally, this study looked for evidence of bias using the relative performance measure of RevPAR index. Given the dominant role of the RevPAR index and its availability for research, this is a good start. However, it might be of use to explore how the inclusion of distant hotels affects profit-based relative performance measures. While these are less commonly used to and therefore are less available for research, the profit-based measures might be of greater interest, especially to owners and investors.

Footnotes

ORCID iD

Timothy Webb

Notes

Zvi Schwartz, PhD (e-mail: zvi@udel.edu), is chairperson and professor of hospitality business management at the Alfred Lerner College of Business and Economics, University of Delaware, Newark, DE. Timothy Webb, PhD (e-mail: twebb@udel.edu), is an assistant professor of hospitality business management at the Alfred Lerner College of Business and Economics, University of Delaware, Newark, DE.

References

Baruca

P. Z.

Čivre

(2012). How do guests choose a hotel? Academica Turistica, 5(1), 75-84.

Buckhiester

(2011). Revenue management as a multi-disciplinary business process: Part two. Journal of Hospitality Financial Management, 19(2), 97-113. https://doi.org/10.1080/10913211.2011.10653916

Butler

(2016, June 14). Why you may need to change your “competitive set” now. STR adopts new guidelines. HotelLAwBlog. https://hotellaw.jmbm.com/may-need-change-competitive-set-now-str-adopts-new-guidelines.html

Chen

M. J.

(1996). Competitor analysis and interfirm rivalry: Toward a theoretical integration. Academy of Management Review, 21(1), 100-134. https://doi.org/10.5465/amr.1996.9602161567

Chipkin

(2007). PWC: ADR, RevPAR growth to continue. Hotel & Motel Management, 222, 4.

Cohen

J. E.

(1988). Statistical power analysis for the behavioral sciences. Lawrence Erlbaum.

Cross

R. G.

Higbie

J. A.

Cross

D. Q.

(2009). Revenue management’s renaissance: A rebirth of the art and science of profitable revenue generation. Cornell Hospitality Quarterly, 50(1), 56-81. https://doi.org/10.1177/1938965508328716

Enz

C. A.

Canina

Walsh

(2001). Hotel-industry averages: An inaccurate tool for measuring performance. Cornell Hotel and Restaurant Administration Quarterly, 42(6), 22-32.

Enz

C. A.

Peiró-Signes

Á.

Segarra-Oña

M. D. V.

(2014). How fast do new hotels ramp up performance? Cornell Hospitality Quarterly, 55(2), 141-151. https://doi.org/10.1177/1938965513506518

10.

Fox

(2011). Price and location continue to dominate hotel booking, agents rarely in the mix. Tnooz. https://www.tnooz.com/article/price-and-location-continue-to-dominate-hotel-booking-agents-rarely-in-the-mix/on7/20/2018

11.

Garen

J. E.

(1994). Executive compensation and principal-agent theory. Journal of Political Economy, 102(6), 1175-1199. https://doi.org/10.1086/261967

12.

Khalilzadeh

Tasci

A. D.

(2017). Large sample size, significance level, and the effect size: Solutions to perils of using big data for academic research. Tourism Management, 62, 89-96. https://doi.org/10.1016/j.tourman.2017.03.026

13.

Kim

J. Y.

Canina

(2011). Competitive sets for lodging properties. Cornell Hospitality Quarterly, 52(1), 20-34. https://doi.org/10.1177/1938965510377606

14.

Laffont

J. J.

Martimort

(2009). The theory of incentives: the principal-agent model. Princeton university press.

15.

Lakens

(2013). Calculating and reporting effect sizes to facilitate cumulative science: A practical primer for t-tests and ANOVAs. Frontiers in Psychology, 4, 863. https://doi.org/10.3389/fpsyg.2013.00863

16.

Love

A. S.

Walker

B. H.

Sutton

D. W.

(2012). New option in hotel appraisals: Quantifying the revenue enhancement value of hotel brands. Appraisal Journal, 80(3), 223-234.

17.

Morgan

M. S.

Dev

C. S.

(1994). Defining competitive sets of hotel brands through analysis of customer brand switching. Journal of Hospitality & Leisure Marketing, 2(2), 57-91. https://doi.org/10.1300/J150v02n02_06

18.

Murphy

K. J.

(1999). Chapter 38: Executive compensation. In Handbook of labor economics (Vol. 3, Part B, pp. 2485-2563). Elsevier. https://doi.org/10.1016/S1573-4463(99)30024-9

19.

Otainsight. (2018, March 21). How to select your perfect competitive set. https://www.otainsight.com/resources/blog/how-to-select-your-perfect-competitive-set

20.

Peteraf

M. A.

Bergen

M. E.

(2003). Scanning dynamic competitive landscapes: A market-based and resource-based framework. Strategic Management Journal, 24(10), 1027-1041. https://doi.org/10.1002/smj.325

21.

Rivers

M. J.

Toh

R. S.

Alaoui

(1991). Frequent-stayer programs: The demographic, behavioral, and attitudinal characteristics of hotel steady sleepers. Journal of Travel Research, 30(2), 41-45. https://doi.org/10.1177/004728759103000209

22.

Schwartz

Altin

Singal

(2017). Performance measures for strategic revenue management: RevPAR versus GOPPAR. Journal of Revenue and Pricing Management, 16(4), 357-375. https://doi.org/10.1057/rpm.2016.23

23.

Selya

A. S.

Rose

J. S.

Dierker

L. C.

Hedeker

Mermelstein

R. J.

(2012). A practical guide to calculating Cohen’s f², a measure of local effect size, from PROC MIXED. Frontiers in Psychology, 3, 111. https://doi.org/10.3389/fpsyg.2012.00111

24.

Slattery

(2002). Reported RevPAR: Unreliable measures, flawed interpretations and the remedy. International Journal of Hospitality Management, 21(2), 135-149. https://doi.org/10.1016/S0278-4319(01)00034-2

25.

Smirnova

A. S.

Zavertiaeva

M. A.

(2017). Which came first, CEO compensation or firm performance? The causality dilemma in European companies. Research in International Business and Finance, 42, 658-673. https://doi.org/10.1016/j.ribaf.2017.07.009

26.

Smith

R. A.

Zheng

(2011). A look at comp sets: A historical perspective that shapes today’s way of thinking. Cornell Hospitality Quarterly, 52(4), 371-373. https://doi.org/10.1177/1938965511420550

27.

Stiglitz

J. E.

(1988). Principal and agent [Discussion paper]. Woodrow Wilson School of Public and International Affairs, Princeton University, Princeton, NJ, United States.

28.

Tsaur

S. H.

Tzeng

G. H.

(1996). Multiattribute decision making analysis for customer preference of tourist hotels. Journal of Travel & Tourism Marketing, 4(4), 55-69. https://doi.org/10.1300/J073v04n04_04

29.

Watkins

(2015, July 7). 6 Secrets to calculating a better comp set. Duettocloud. http://duettocloud.com/6-secrets-to-calculating-a-better-comp-set/

30.

Watkins

(2017, May 18). In uncertain times, focus on RevPAR index as well as RevPAR growth. Duetto Digest. http://duettocloud.com/uncertain-times-focus-revpar-index-well-revpar-growth/

31.

Webb

Schwartz

(2017). Revenue management analysis with competitive sets: Vulnerability and a challenge to strategic co-opetition among hotels. Tourism Economics, 23(6), 1206-1219. https://doi.org/10.1177/1354816616671473

32.

Wong

K. K.

Chi-Yung

(2002). Predicting hotel choice decisions and segmenting hotel consumers: A comparative assessment of a recent consumer based approach. Journal of Travel & Tourism Marketing, 11(1), 17-33. https://doi.org/10.1300/J073v11n01_02