Developing the Full Picture on Hotel Industry Averages, by Cathy A. Enz, Ph.D., Linda Canina, Ph.D., and Kate Walsh, Ph.D. (FULL TEXT VERSION)
Hotel operators and observers often employ industry-wide averages as key points of comparison and analysis for room rates, occupancy, and revenues. The use of simple averages, however, can be misleading if one does not take into account the possibility that a mean will be pulled in one direction or another by extreme values. This analysis of three industry averages shows that those averages are, indeed, subject to distortion, or skew.
A more complete picture of the industry's ADR, RevPAR, and occupancy is gained by examining two other measures: the median, which is a measure of the data's middle value, and the mode, which states the most common data point. By comparing the mean with the median and the mode, one can determine the extent to which the mean overstates the industry's ADR and RevPAR and understates the typical occupancy. Specifically, 61 percent of U.S. hotels recorded a RevPAR below the overall mean and 63 percent saw an ADR below the mean, but only 48 percent reported occupancy below the mean.
Many of the extreme values are found in the top-25 markets, which have hotels with inordinately high ADRs . Analysis of those markets shows that, once again, the overall statistics are distorted by a relatively small set of hotels with exceptional ADRs and occupancies. However, each of the top markets shows a distinctive rate and occupancy pattern.
The pattern of skewed operating statistics carries over into individual lodging segments. The greatest distortions arise in the luxury and upscale segments, while economy and budget hotels record more consistent (normally distributed) statistics.
Finally, the analysis shows that although the events of September 11, 2001, created much turmoil for the industry, the hotel business had already cooled substantially from its record pace of a year earlier.
In conclusion, managers must be careful in applying overall industry statistics to their own situation and should take into account the factors that distort operating statistics.
HOTEL OPERATORS AND INVESTORS USE A number of industry statistics as benchmarks to assess current operations and to make forecasts and plans. Three commonly used statistics are occupancy, average daily rate (ADR), and revenue per available room (RevPAR). ADR is the average daily rate per rented room, or the mean price charged for all hotel rooms sold in a given period. RevPAR is calculated by dividing revenue by the number of rooms available for sale. Occupancy is calculated by dividing the number of rooms sold by the number of rooms available and multiplying by 100. Most often summarized as means (averages), these three measures have become common benchmarks by which the lodging industry makes performance comparisons.
By analyzing and interpreting the key industry averages of ADR, RevPAR, and occupancy, it is possible to understand the nature of those averages. In particular, it is worth assessing whether and the extent to which the means of those statistics may mask variations in the pattern of industry demand. This article reports our analyses of the nature of the industry averages commonly in use. It also explains the usefulness and accuracy of two other statistics, namely, the median and modal values.
The study is based on monthly demand data for brand-name hotels in the United States for the 12-year period between 1988 and 2000, followed by additional analysis on a daily basis for the months of September 2000 and September 2001. The data were drawn from the
Smith Travel Research database, which is effectively a census of U.S. brand hotels. This comprehensive sample captures over 98 percent of the brand-hotel inventory. Thus, it is widely considered to be fully representative of the entire U.S. lodging population for brand hotels.
Most people are familiar with the concept of the mean, or average, which gives the value of the central tendency or location of data. The arithmetic mean is a single number that gives the central location of a set of data.1 The mean is the value obtained when one divides the sum of all values in the data set by the number of items in the set. A chief problem with the arithmetic mean is its tendency to be distorted by extreme values at either end of a distribution. Indeed, the mean can be so affected by extreme values that it is pulled in the direction of these extremes. As we demonstrate below, this occurs in the hotel industry. If a major market such as New York City or a group of key markets has extremely high ADR figures (as New York and certain others do), they can pull the entire U.S. average ADR in this positive direction. In such a situation, the mean may not be representative of the typical ADR in the industry overall, and thus it masks the industry's common ADR patterns.
When the arithmetic mean is influenced by extreme values, two other statistics that are measures of central tendency, the median and the mode, may be useful for analysis and comparison. The median captures the most central or middle value. Put another way, the median is the value below and above which lay an equal number of data points. Unlike the mean, it is the middlemost value in a set of numbers. Since it is based more on the size of the sample than on the numeric values, it has the benefit of being relatively free from the distortion experienced by the mean when a distribution contains extreme values. The median tells us that 50 percent of branded hotels are above and 50 percent of these hotels are below a particular number. In our study, for instance, the average ADR for branded hotels over the 12-year period was $63.43, while the median for the same period was $7.43 less, at $56.00. The mode shows the value that occurs with the most frequency. Unlike the median and the mean, the mode always appears as a value in the data set. It is often called the most fashionable value (i.e., à la mode), because it captures the most typical or representative value located where the data have maximum cluster. Modes for the three performance measures in this study show the industry's most common RevPAR, ADR, and occupancy values. Returning to overall industry ADRs, the U.S. mode is $47.00, or $16.43 less than the industry mean. Thus, typical hotels in the U.S. were reporting ADRs below the average during the 12-year period of our study. By including the median and the mode in our analytical toolkit, we can get a fuller and somewhat different picture of the central tendency of lodging performance. With these three measures of central tendency in hand we turn to our study to examine the pattern of frequency distributions in lodging demand since 1988.
Exhibit 1 shows the overall mean, median, and mode for each of the three key elements of demand, RevPAR, ADR, and occupancy during the 12-year period. All of the data were adjusted to year- 2000 dollars using yearly consumer- price-index values to control for inflation. We used a total of 1.8 million observations to calculate our statistics.2 As the table shows, the industry average RevPAR and ADR are both higher than the typical (mode) or middle (median) points of those hotel-performance measures.
At $41.55, the mean RevPAR is $11.55 higher than either the median or mode, while the ADR value of $63.43 is $16.43 higher than the modal or median values. By looking at the distribution of brand hotels' RevPARs and ADRs we see that 61 percent reported RevPARs below the industry average, and 63 percent recorded ADRs below the industry average. If one relied solely on the average, one might conclude that most hotels enjoy numbers close to the mean, but far fewer hotels hit that average. Instead, a smaller number of hotels that exceed the average are, in fact, substantially above the average. A look at occupancies shows the opposite pattern from that of ADR and RevPAR. The typical hotel occupancy (mode) of 70 percent is higher than the average of 63 percent. The distribution reveals that 52 percent of U.S. hotels experience higher occupancy than the average value.
What causes this is that the extremely low occupancy of some hotels pulls the industry average down to a figure that is 7-percent lower than the typical occupancy levels (mode) and 2-percent lower than the middle point (median). The substantial differences among the average (mean), median, and mode suggest that reliance on just the average could be misleading. Over this period of time, the industry averages for ADR and RevPAR overstated the performance of the typical hotel. Those averages were skewed by a small group of hotels or markets that have much higher ADRs and RevPARs than do all the rest. By the same token, typical hotel occupancies have been understated by reliance on an average that is distorted by extremely low occupancies of a small group of hotels or markets.
If the mean, median, and mode for the hotel industry's ADRs, RevPARs, and occupancies were all the same value or close to the same number we would be able to conclude that the hotel industry experienced what is known as a normal distribution for these statistics. A normal distribution is commonly known as a bell curve, which is a reasonably symmetrical distribution of the data. That does not occur in our hotel-industry data, however, and instead lodging demand displays a non-symmetrical, or skewed, distribution. This skewness is a key factor in producing the inconsistencies that we have demonstrated with regard to the lodging-performance benchmarks. Skewness results when the frequency distribution has a heavy mass of extreme values. A symptom of skewness ccurs when the extreme values pull the mean away from the median and the mode in the direction of the extreme values. The median is also pulled away from the mode in the same direction as the average, but the median is not affected as much by extremes as the mean is. This skewed pattern of lodging demand is shown in Exhibit 2. Plotting the frequency distribution of hotels' ADR, RevPAR, and occupancy numbers for each month over the 12-year period reveals the shape of the distribution. The mode is the highest point on the curve, which is the area where the largest number of hotel values cluster. When the mean is less than the median, as it is for occupancy, the distribution is said to have a negative skewness (i.e., skewed to the left). In contrast, for RevPAR and ADR the mean is greater than the median and mode, and hence the distribution has a positive skewness (i.e., skewed to the right). The position of the mean, median, and mode in these graphs reveals the distance between these statistics and shows the asymmetry in U.S. lodging demand. Given the above results for overall demand, hospitality managers may have one of two reactions. The first is that considering the differences in the values of the mean, median, and the mode (in addition to the wide variability of these demand statistics), using overall U.S. lodging demand averages alone may prove to be far too simplistic. The second is that they already ignore overall averages and focus their analysis on the local market's lodging demand data. While analysts and chainwide brand-pricing strategies may be more focused on the overall industry measures, many propertylevel managers and consultants doing feasibility studies use their own markets and competitive sets to set performance benchmarks.
Nevertheless, decision makers may not fully grasp the variability within markets. As such, it is important to understand the patterns of differences that may be due to market location or segment. To refine our understanding of the patterns of lodging demand we now turn to an exploration of key markets and market segments.
Our analysis revealed that substantial variation exists in overall lodging demand. To determine whether this variation is the result of distinctive patterns in specific markets, we explored whether primary markets (which we defined as the top-25 markets) showed significantly different demand patterns from the remainder. By conducting a series of statistical tests to determine whether the arithmetic means for the primary markets were similar to those of the remaining markets, we found that the major markets have significantly different performance levels. By splitting the
branded hotels into two groups, one composed of the top-25 markets and the other consisting of all other U.S. markets, we clearly see the difference in their demand patterns.3
The findings reported in Exhibit 3 show that the RevPARs and ADRs in both primary and other markets are higher than the typical hotels' performance levels. The degree of difference between the average and the typical hotel was substantially greater for the top-25 markets than for the others. In these markets, the typical hotel's RevPAR of $30.00 was $22.56 below the average RevPAR. In contrast, the typical RevPAR of $26.00 in other markets was only $11.73 lower than the average. The big gap between typical and average hotel for the top-25 markets suggests that the average is unduly influenced by the extremely high RevPARs of a relatively small set of hotels. The smaller gap between the typical and average hotel in all other markets suggests that the bulk of U.S. markets have more normal distributions without xtraordinarily high RevPARs.
Our exploration of ADRs revealed a similar but more pronounced pattern. For the top-25 markets the gap between the typical hotel and the average hotel was $24.35, while the gap was $11.93 for ADRs in other markets. All other markets were less volatile than the top-25 markets, indicating a narrower range of demand values. ADRs for some markets are so high in the top-25 markets that they again unduly distort the average and push it to a higher level than the typical hotel ADRs. In both the case of ADRs and RevPARs the average overstates performance when compared to the typical hotels. Average occupancy figures were not as dramatically different between primary and other markets. While occupancies were 4.5- percent higher in the primary markets, the averages were consistently lower than the typical hotel occupancies in both types of markets. The primary markets average occupancy of 66.26 percent was 8.64-percent lower than the modal hotels, while the average occupancy in all other markets (at 61.87 percent) was 8.13-percent lower than the typical hotels in those markets. Again, the results reveal negative skew such that the average occupancy is lower than the typical hotel occupancy. So, the average understates the level of rooms inventory that is efficiently filled with guests. Our market comparisons showed significant differences between primary and other markets, but the occupancy differences were smaller than for RevPAR or ADR. In addition, the elevated levels of demand in some markets may overstate demand for secondary markets, but that may also be true for other top-25 markets. The degree of variation in demand was found to be so great in the top-25 markets that treating them as though they are a group may be problematic. Since these markets have great variability in their performance, we next turned to an examination of key cities within the top-25 markets to more fully understand the degree to which variability in demand patterns exist.
While our study showed that the top-25 markets were distinct from secondary markets, the high volatility and skewness in demand would suggest that key cities should be examined separately to see how similar they are to each other. We chose ten key cities for this analysis. They are Atlanta, Boston, Chicago, Los Angeles, Miami, New York, Philadelphia, Phoenix, San Francisco, and Washington, D.C. We chose RevPAR as the performance indicator for these tests because it is often considered the most critical measure of operating performance, and by definition, encompasses an element of rate and room supply.4 Conducting a series of statistical tests, we explored the differences between the mean RevPAR of each of the ten cities compared with the mean of the top-25 markets excluding that city. Based on this analysis we found each of the ten key cities to have RevPARs significantly different from the RevPARs of the remaining top-24 markets. Thus, each top-ten city is distinctive from the remaining top markets. Exhibit 4 shows the means, medians, and modes for each of the ten cities. In analyzing differences between a particular city and the remaining top-25 markets, we found that New York City has the highest degree of variability, as well as the largest gap between the average of all hotels and modal value that represents the typical hotel. (That difference was $48.10.) The table lists the cities in the order of the gap between their mean and their mode, showing that San Francisco, Phoenix, Los Angeles, and Boston are the cities with the next-highest degrees of difference between the average and typical hotel. Interestingly, Los Angeles, Phoenix, and Atlanta have the lowest average RevPARs, while Miami has the smallest gap in the average and the typical hotel RevPAR levels.
The frequency distributions of RevPAR for these ten cities are shown in Exhibit 5 (next page). New York City has the highest degree of variability in RevPAR, while Atlanta has the lowest. In fact, the RevPAR pattern for NYC is unlike that of any other city— owing both to its wide range of values and its extremely high values. The RevPAR patterns of Boston, Chicago, Philadelphia, San Francisco, and washington, D.C., are reasonably similar. It is interesting to note that we have heard Washington, D.C., being discussed as though its demand pattern were like that of NYC, when in fact the D.C. market's demand patterns are closer to those of Chicago and Philadelphia. Even given the similarities of the five markets we just discussed, the statistical tests and graph of distributions reveal that RevPAR demand in the key cities is not comparable. Moreover, NYC's RevPAR inflates the average substantially for the top-25 markets.
Another way to examine the variability in demand patterns is to examine the differences by price segment—to see whether demand in some segments is skewed and to see the extent to which demand patterns vary from segment to segment. To investigate these matters we looked at the means, medians, and modes for the following five price segments: luxury, upscale, midprice, economy, and budget. In our analysis of price segments, presented in Exhibit 6, we see that the same pattern exists in each segment (similar to the overall industry). That is, for each segment the mean exceeds both the median and the mode for both RevPAR and ADR, while the reverse relationship holds for occupancy. Occupancy is negatively skewed, given that the mean is lower than the median and the mode. While the typical hotel has lower RevPAR and ADR figures than the average, the differences in these statistics are greatest for luxury hotels, with a $22.34 difference between the average ADR and the typical hotel ADR. Upscale hotels have only a $12.59 difference, with midprice, economy, and budget segments reporting differences of between $7.69 and $3.51 when comparing the average with the typical hotel in their price segments.
While luxury hotels have the greatest degree of variation in
ADRs and RevPARs, this segment has less variation in occupancy levels. Budget and economy properties have more normalized distributions (less variation and skew in their performance patterns). Thus, industry averages are more useful for individuals who focus on these two price segments. Clearly, caution is warranted for those who want to understand performance in upscale hotels. To a greater extent than other segments, care is necessary when interpreting performance based on averages in luxury hotels.
On September 11, 2001, the United States experienced terrorist attacks on the cities of New York and Washington, D.C., that profoundly changed the existing economic activity in the lodging industry. To explore the implications of these events on one's ability to make sense of industry averages we broke the month of September into three sections. The first period, September 1 through 10, includes the Labor Day holiday, which produces low ADRs and occupancies. (In fact, the lowest occupancy for September was reported on September 3.) The second period was September 11 through 16, the week after the terrorist attacks, during which the fact that people were stranded in key cities may have artificially inflated occupancy. The last period was September 17 through 29, during which the wide perception was that people were "not traveling." For this analysis we relied on September's daily occupancy data for the years 2000 and 2001. By calculating the means, medians, and modes for the total U.S. and for New York City and three other key cities most directly affected by the attack (Boston, San Francisco, and Washington, D.C.), we can see a picture of huge variability that is illuminated by looking at both the average and the typical hotel. As the performance measures in Exhibit 7 show, New York City and the three other key cities saw substantial declines in occupancy. Before the events of September 11, New York occupancies were 7.42-percent lower than the banner year of 2000. Immediately after the tragedy year-to-year occupancies were down by 20.12 percent between September 11 and 16, and went further down in the third period (September 17 to 29) to 22.33 percent from the previous year. Drops in occupancy of the three key cities were even larger than in NYC.
Looking at the U.S. overall, we see a 2.79- percent occupancy drop in 2001 from 2000 before September 11, with year-to-year drops of 10.10 percent in the week after the attack, and 9.27 percent in the final September period. While these figures are strong indicators of a decline, the modal or typical hotel's performance shows a far more dramatic drop than the averages reveal.
Exhibit 8 shows the distribution of occupancy levels for the three time periods. What is notable about these figures is that they show that the typical hotel's occupancy, which had been at the 90-percent level, was cut nearly in half, to around 50 percent. The dramatic shift in the typical hotel's occupancy, rather than the average occupancy, has led many in the field to express alarm. On the other hand, those looking at overall averages may conclude that a far less dramatic crisis has occurred. These statistics and figures show how important it is to consider more than one statistic when attempting to make forecasts of economic impact and future recovery.
The fundamental message our study offers is to proceed with caution when using industry averages for forecasting and making decisions. Unstable and turbulent environments can produce extreme values that pull the arithmetic mean in one direction or another. As our study has shown, that average alone is not adequate to describe lodging demand fully, particularly since the industry reflects such large variation in markets, key cities, and price segments. Reliance on the average can lead managers to both overstate
ADR and RevPAR goals and understate occupancy goals. We now know that the average overstates ADR and RevPAR because
of extremely high numbers in some segments (particularly, luxury) and certain markets (notably, New York City). Occupancies are actually higher for the typical hotel than the averages would suggest.
We have shown that the "average" masks the true patterns of lodging demand, as measured by RevPAR, ADR, and ccupancy, and hence we caution managers to be careful not to make decisions on the basis of easily available data, or convenient reference points. Consciously or unconsciously managers may insufficiently adjust up or down on the basis of the averages, and hence prevent their final judgments from being as informed and reasonable as they could.
To ensure that one uses the appropriate demand data to compare against operating performance and to prepare forecasts, we advise augmenting aggregate industry measures with additional statistics, such as the median and mode. Another statistic, the standard deviation, which expresses the extent to which the data are dispersed, can also provide useful information, although we did not discuss it here. Using all these statistics may enable decision makers to more fully explore the possible role of skewness in lodging
demand and identify areas where current measures may be distorted.
Our results suggest that key cities dominate and distort average lodging demand for brand hotels. Overall lodging data mask huge variability that exists by market and segment. In addition, performance patterns vary substantially within markets and segments. A look at daily demand in light of the events of September 2001 further supports the observation that the lodging industry has patterns of demand in key cities that shape our impressions of the entire U.S. market. It makes sense to add measures of performance to the analysis, the better to develop a fuller understanding and more appropriate comparison points for meaningful benchmarking. These additions would enable managers and investors to make more effective analyses and more precise decisions. In addition, selecting the right comparison or reference city, segment, or market for a benchmark will go a long way toward improving the hospitality decision maker's toolkit.
CHR Reports
This report is produced for the benefit of the hospitality industry by The Center for Hospitality Research at the School of Hotel Administration.
Cathy A. Enz, Executive Director - Glenn Withiam, Director of Publications Services - Copyright © 2002 by Cornell University
The CHR's mission is to foster and conduct research of significance to the hospitality industry.
At The Center for Hospitality Research we:
- Conduct research on issues of high priority to the industry;
- Support research fellows who rigorously investigate important topics;
- Broadly distribute research findings to enhance management practice; and
- Build relationships with the industry to enhance a knowledge community.
Visit us at
For more information or to become a Partner of The Center for Hospitality Research, contact us at
Shaping the global knowledge base for hospitality management