Abstract

Past research has shown that individuals vary in their attitudes and behaviors regarding weather forecast information. To deepen knowledge about these variations, this article explores 1) patterns in people’s sources, uses, and perceptions of everyday weather forecasts; and 2) relationships among people’s sources, uses, and perceptions of forecasts, their personal characteristics, and their experiences with weather and weather forecasts. It does so by performing factor and regression analysis on data from a nationwide survey of the U.S. public, combined with other data. Forecast uses factored into planning for leisure activities and for work/school-related activities, while knowing what the weather will be like and planning how to dress remained separate. Forecast parameters factored into importance of precipitation parameters and of temperature-related parameters, suggesting that these represent conceptually different constructs. Regression analysis showed that the primary drivers for how often people obtain forecasts are what they use forecasts for and their perceived importance of and confidence in forecast information. People’s forecast uses are explained in large part by their frequency of obtaining forecasts and their perceived importance of temperature-related and precipitation forecast information. This suggests that that individuals’ frequency of obtaining forecasts, forecast use, and importance of forecast parameters are closely interrelated. Sociodemographic characteristics and, to a lesser extent, weather-related experience also influence some aspects of people’s forecast sources, uses, and perceptions. These findings continue to build understanding of variations among weather forecast users, which can help weather information providers improve communication of forecasts to better meet users’ needs.

1. Introduction

Weather forecasts are a common part of people’s lives in the United States, with millions of individuals obtaining forecasts daily and using them in a variety of decisions (Lazo et al. 2009, hereafter LMD09). Recent surveys find that approximately half of the U.S. public reports following weather news “very closely,” with “no other topic generat[ing] close to this level of interest” (Pew Research Center 2008, p. 39). Weather information providers offer a variety of forecast content and formats through multiple communication channels, creating a complex, dynamic information environment. Further, advances in technology, media evolution, and provider responses to user needs are rapidly changing the landscape of available information. Given the widespread provision and use of everyday weather forecasts, we have surprisingly little knowledge about how and why members of the U.S. public obtain, use, and perceive this information. From a communication and use perspective, building this knowledge is fundamentally interesting. Because new forecast products are continually being developed, this knowledge is also an important contributor to improving communication of weather forecasts to meet users’ needs, as recommended by several recent multidisciplinary working groups (e.g., NRC 2006, 2010; Morss et al. 2008b).

People’s attitudes and behaviors regarding everyday weather forecast information have been investigated in several previous studies, both for the U.S. public in general (e.g., Harris Poll 2007; LMD09) and for specific subpopulations (e.g., CFI Group 2005). For example, LMD09 examined people’s sources, perceptions, and uses of weather forecast information using data from a nationwide survey of the U.S. public. A Harris poll (Harris Poll 2007) examined people’s sources, perceptions, and uses based on a priori categorizations of gender, age cohort (i.e., generation), and geographic region. Results from these efforts indicate that, in general, individuals obtain, use, and perceive weather forecast information in diverse ways. Related work has also been conducted in the field of mass communication. For example, Tan (1976) examined people’s media uses and preferences for obtaining weather forecasts, but these results are in need of updating, and most mass communication studies focus on other information domains.

These studies have offered an important start, but they have been largely descriptive; they have not examined the differences among individuals’ attitudes and behaviors or the underlying reasons in depth. The analysis presented here seeks to extend the existing knowledge in this area by investigating 1) patterns in people’s sources, uses, and perceptions of forecasts; and 2) relationships among people’s sources, uses, and perceptions of forecasts, their personal characteristics, and their experiences with weather and weather forecasts. We focus on sources, uses, and perceptions because we broadly conceptualize them as core aspects of how members of the general public interact with weather forecasts. Through our analysis, we aim to begin organizing and explaining similarities and differences in people’s sources, uses, and perceptions of forecasts. Given the intricacies of this process and the limited previous work in this area, the current article takes an exploratory approach. Our goal is to begin developing an empirical understanding that can be built on in future work and eventually used to extend or develop theory. To our knowledge, no other study has undertaken this type of analysis about weather data on a nationwide scale.

To begin to assess these issues, we employ statistical analysis across a broad set of relevant concepts. Doing so allows us to investigate patterns and relationships that emerge empirically rather than imposing a priori stratifications. For sources and uses, we examine individuals’ frequency of obtaining forecasts from different sources and their use of forecasts for different activities, respectively, using data from questions on the LMD09 survey. For perceptions, we examine individuals’ stated importance of different forecast components, importance of National Weather Service (NWS) forecast information, forecast confidence, and overall forecast satisfaction, again using LMD09 survey data. To begin exploring the roles of the forecasts and weather where individuals live, we analyze the sources, uses, and perceptions data in conjunction with data on weather variability and NWS forecast accuracy matched to survey respondents’ locations. The analysis also includes data from the survey on respondents’ weather-related exposure and their sociodemographic characteristics.

As noted above, there is considerable diversity in the sources available for obtaining forecasts, activities for which people use forecasts, and physical parameters provided in forecasts. Thus, our first set of analyses examines underlying commonalities and differences by conducting factor analysis to explore whether there are latent groups—also known as factors—of people’s sources, uses, and perceptions of weather forecast information. From a theoretical perspective, understanding latent factors can help develop deeper knowledge about attitudes and behaviors that can be compared, applied, or generalized across other situations. From a practical perspective, understanding these factors can help identify audience differences and their characteristics, which can help weather information providers improve communication to meet users’ needs.

We then conducted linear regression analyses to explore how people’s frequency of obtaining forecasts, uses of forecasts, and importance of forecast parameters are influenced by their attitudes, behaviors, characteristics, and experiences. We intentionally examined people’s forecast-related attitudes and behaviors from a broad perspective, analyzing relationships among multiple concepts at the same time. To do so, we chose to include measures representing a range of concepts as independent variables in the regressions, rather than studying the influence of a single concept (e.g., forecast accuracy) in detail. We chose to not make a priori assumptions about the directions of relationships among people’s sources, uses, and perceptions because these relationships can be complex and multidirectional. For example, one’s source of a forecast may affect one’s use of that forecast, which may then affect from where one obtains forecasts in the future. Thus, as a first investigation, we included measures of sources, uses, and perceptions as independent variables in some regressions as well as dependent variables in other regressions.

People regularly experience weather and receive weather forecasts and, as they do so, they form impressions about the characteristics of weather and the forecast accuracy where they live (Morss et al. 2008a). These impressions about weather and weather forecasts may then serve as important mediating influences on their future attitudes and behaviors. To begin exploring the influence of different types of experience on people’s attitudes and behaviors, we included weather exposure, forecast accuracy, and weather variability as independent variables in the regressions. To improve understanding of how individuals’ personal characteristics influence their attitudes and behaviors, we also included sociodemographic characteristics as independent variables in the regressions. By analyzing these variables together, we can begin to form a broad picture of how and why people obtain, use, and perceive forecasts as they do.

In addition to providing a cross-sectional look at variations in people’s attitudes and behaviors regarding weather forecast information, this analysis can serve as a baseline for future comparative analyses to identify changes over time. The methodological approaches used in this study could also be applied to examine specific types of weather forecasts or events, climate forecast information, or specific users of forecast information.

The next section briefly describes the data used in this study. Sections 3 and 4 discuss the factor and regression analyses, respectively, and their results. The final section summarizes the key findings and discusses areas for future research.

2. Data

The data used in our analyses came from three sources—a nationwide survey of the U.S. public, NWS forecast verification data, and National Climatic Data Center (NCDC) weather observation data.

a. Data from a nationwide survey of U.S. public

The data on people’s sources, uses, and perceptions of forecasts, weather-related exposure, and sociodemographic characteristics came from a nationwide, Internet-based survey of the U.S. adult public implemented by the authors. Other analyses with these data and those from other survey questions are reported in Morss et al. (2008a), LMD09, and Morss et al. (2010). The survey questions examined here, except for the sociodemographic questions, are reproduced in the appendix; the full survey is available from the authors.

We designed and implemented the survey using accepted survey research methods (e.g., Dillman 2000; Tourangeau et al. 2000; Presser 2004). Starting from and building on survey questions used in previous research by Lazo and Chestnut (2002), we developed the survey questions iteratively with colleagues based on our and the meteorological community’s interest in people’s sources, uses, and perceptions of forecasts and related concepts. We then pretested the survey in person with nonmeteorologists using verbal protocols (“think-alouds”), which we used to finalize the survey content. We collected the survey data in November 2006. A survey research company (ResearchExec) programmed and hosted the survey and collected the data. A second company [Survey Sampling International (SSI)] provided the sample. SSI e-mailed and invited a large number of people from their U.S. Internet panel to respond to a survey. We checked and confirmed survey functionality and data quality after approximately 100 responses and then continued data collection, accepting the first 1200 responses. Caucasians were overrepresented in this group, so we purposively sampled approximately an additional 300 non-Caucasians. Only people invited from the sample could access the survey, and people were allowed to complete the survey only once.

Given the sampling strategy and the sampling error inherent in Internet-based surveys, the survey was not intended to obtain a truly representative sample or to provide results generalizable to the U.S. public. Nevertheless, the methodology does provide results that are more indicative of the views of members of the U.S. public at large than other commonly used methodologies, such as questionnaires given to students or posted on weather-related web sites. Conducting similar surveys with more representative samples would require additional resources and is a goal of future work. As an indication of how well our survey respondent population corresponds to the U.S. adult public, we compared characteristics of the two groups using data from the 2006 American Community Survey (U.S. Census 2007). Our respondent population has a similar gender and race distribution to the U.S. public, but it is slightly older and more educated, and it somewhat underrepresents people with very low and very high incomes [see Morss et al. (2008a) for details]. Our sample includes respondents from every U.S. state.

We received 1520 completed responses to the survey, but 55 respondents indicated they never use forecast information and so were not asked many of the survey questions examined here (LMD09). Of the remaining 1465 respondents, four could not be geographically matched with weather forecast accuracy or weather variability data (discussed below). Thus, the number of observations in the analysis presented here is n = 1461.

The data on people’s sources, uses, and perceptions of forecasts came from questions about people’s frequency of obtaining weather forecasts from 10 different sources (Q2); how often people use weather forecasts for eight different activities (Q5); how important each of 14 different forecast parameters is to people (Q6); people’s overall satisfaction with weather forecast information (Q7); and people’s confidence in forecasts of six lead times (Q11) (appendix). Following LMD09, we recoded the source data to provide a quantitative, lower-bound frequency of obtaining forecasts by source per month. As a first test of the influence of forecast confidence relative to the other variables tested in the regressions, we summed respondents’ confidence at different lead times to derive a single index of aggregate confidence in weather forecasts.

The data on people’s weather-related exposure came from questions about the average annual percent of on-the-job time people spend outdoors (Q23); the average weekly number of hours people spend traveling to/from work or school (Q24); the average annual percent of leisure time people spend outdoors (Q25); and the average weekly number of hours people spend working outside around home (Q26) (see the appendix).

Sociodemographic data came from survey questions on the number of years respondents had lived within 50 miles of their current residence, gender, age, education level, income level, employment status, and race. We recoded the education data to represent the total number of years of schooling. As a first test of the influence of income, we recoded the income data to the midpoint of the response option ranges. As first tests of the influence of employment and race, we recoded these data to dichotomous variables (i.e., employed full time or not, Caucasian or non-Caucasian).

As needed, we quality controlled quantitative data to replace values that were impossible for variables with physical limits (e.g., age, hours worked per week) with upper bounds of feasible data. Although this meant some potentially unrealistic values were retained in the data, removing this data would have required selecting an arbitrary cutoff. These represent very few cases and have no substantive impacts on the quantitative analysis.

b. NWS forecast verification data

As an approximate representation of the accuracy of weather forecasts in the area where each respondent lives, we used measures of forecast accuracy from NWS verification data. We chose to use these data because the NWS has a standardized method for calculating verification statistics, the NWS data are nationwide and publicly available, and most weather providers use data and techniques similar to NWS’s.

We selected error statistics from those available in the NWS Public Verification Point Forecast Matrix verification dataset (NWS 2009), one for measuring accuracy of maximum temperature forecasts and one for chance of precipitation forecasts. We chose these two forecast parameters because LMD09 show they are highly important to people. For temperature forecast error, we used the root-mean-square error (RMSE) of the one-day maximum temperature forecast, where higher values represent greater forecast error. For precipitation forecast error, we used the Brier score of the one-day probability of precipitation (PoP). The Brier score measures the mean squared probability error based on the forecast probability of an event and whether or not the event occurred (Toth et al. 2003). Brier scores range from 0 to 1, with higher scores representing greater forecast error.

Many people travel near their area of residence (e.g., to go to work, run errands, attend social activities) on a given day and thus experience weather and weather forecasts over a spatial area beyond their zip code of residence. Thus, we chose to use error statistics averaged over an area, the NWS county warning area (CWA). (The CWA is the group of counties for which an NWS Forecast Office is responsible for issuing forecasts and warnings; see http://www.weather.gov/organization.php#maps.) Statistics were averaged over a period from January 2004, when the data were first available, through April 2008. We chose to use a multiyear period to capture seasonal and interannual variability. We matched each survey respondent to a CWA based on reported zip code and then assigned each to the corresponding temperature and precipitation error measures.

c. NCDC weather observation data

To represent the weather variability where people live, we calculated weather variability measures using observations of maximum temperature and precipitation amount from the NCDC Global Surface Summary of the Day (GSOD) dataset. Following the approach of Karl et al. (1995), we defined variability as the mean absolute values of 24-h differences in maximum temperature and in precipitation amount. We used data over the same time period as the verification data—January 2004 through April 2008—for consistency and, again, to capture seasonal and interannual variability.

Temperature and precipitation variability measures were averaged at the NWS CWA level. As with the forecast verification data, each survey respondent was matched to a CWA and corresponding measures based on his/her reported zip code.

3. Factor analysis of sources, uses, and perceptions of forecast information

Previous research has shown that individuals vary in their access of forecasts from different sources, uses of forecasts for different activities, and stated importance of different forecast parameters (LMD09; Lazo and Chestnut 2002). Here we conduct exploratory factor analysis (e.g., Hatcher 2007; Garson 2009a) to investigate whether underlying patterns emerge regarding people’s sources, uses, and perceptions of forecast information. Factor analysis assesses the extent to which two or more items are interrelated because of some shared underlying feature. The resultant factors represent latent constructs that cannot be measured directly, and different factors represent conceptually different constructs.

By analyzing the data for underlying factors without making a priori assumptions about how constructs interrelate, we seek to begin building a fundamental picture of patterns in individuals’ attitudes and behaviors regarding weather information. While the LMD09 survey questions, and thus our data, do not explore all possible sources, uses, and perceptions in depth, examining patterns in these data is a first step toward building this picture, and it indicates some concepts to be explored in more detail in future work. Measuring where individuals fall along factor scales also provides information on respondent heterogeneity that we explore further in the regression analysis (section 4).

We performed1 the factor analysis using the principle axis factoring extraction method with an orthogonal varimax rotation. To select criteria for the factor analysis, we followed guidance from Hatcher (2007) and Garson (2009a) with the goals of our study in mind. We retained factors that accounted for at least 10% of the variance in the dataset, required variables to have a minimum value of 0.4 to load onto a factor, and required the final set of variables loading onto a factor to have reliability (Cronbach’s α) of 0.6 or greater. We also assessed the factors for face validity, checking that variables loading onto a given factor seemed conceptually similar and those loading onto different factors seemed to represent different constructs.

The factor analysis on the sources of forecast information revealed no factors that met all criteria. Although we anticipated there would be similarities and differences among individuals’ uses of the sources that would lead to different groups, the analysis did not robustly support this. Among our survey respondents, 83.5% reported using three or more sources at least weekly, and 43.6% use five or more sources weekly. That no factors emerged indicates that individuals use substantially different combinations of sources. In other words, there is not a sufficiently common set of source usage patterns by respondents to generate a signal in our data.

The factor analysis on forecast uses revealed two factors (Table 1; Fig. 1). Four variables loaded onto factor U1 (α = 0.77), all of which represent personal activities that (most) people do not engage in daily. Thus, we labeled this factor “Forecast use for planning leisure activities.” Two variables loaded onto factor U2 (α = 0.61), which we labeled as “Forecast use for planning work/school-related activities.” Although it is desirable to have more than two variables load onto a factor, it is not uncommon to retain a factor with only two variables in exploratory research (e.g., Kaye and Johnson 2002; Papacharissi and Mendelson 2007). These factor analysis results empirically support our expectation that there are behavioral differences in how people use forecasts. They begin to reveal what different types of forecast uses exist—that is, for leisure and work/school.

Table 1.

Results of factor analysis on uses of weather forecasts. Factor loadings are shown for the uses that load onto a factor (N = 1461).

Results of factor analysis on uses of weather forecasts. Factor loadings are shown for the uses that load onto a factor (N = 1461).
Results of factor analysis on uses of weather forecasts. Factor loadings are shown for the uses that load onto a factor (N = 1461).
Fig. 1.

Factor analysis on uses of weather forecasts. The variables planning how to dress yourself or your children and simply knowing what the weather will be like did not load onto a factor, so these items were considered independently.

Fig. 1.

Factor analysis on uses of weather forecasts. The variables planning how to dress yourself or your children and simply knowing what the weather will be like did not load onto a factor, so these items were considered independently.

The two remaining use items—“simply knowing what the weather will be like” and “planning how to dress yourself or your children”—did not load onto a factor. This suggests that these items are distinct from each other and that they may represent forecast use constructs other than leisure and work/school. For instance, “simply knowing what the weather will be like” could represent general use of weather forecasts (i.e., not for specific activities), information surveillance (i.e., to keep up with the weather), entertainment, or some other construct. Related motivations for media use have been identified in uses and gratifications research from the field of mass communication, investigating other information domains (e.g., Lasswell 1948; Papacharissi and Mendelson 2007; Papacharissi and Rubin 2000). Future research efforts could apply these concepts from uses and gratifications theory to the weather domain to help understand people’s motives for getting forecasts. “Planning how to dress yourself or your children” may also reflect a broader construct, such as forecast use for regular, daily activities. Future surveys would need to test additional items to show whether or not these and/or different constructs exist. Because “simply knowing what the weather will be like” and “planning how to dress yourself or your children” are the top two forecast uses of those tested in our survey, with 72% and 55% of respondents usually or always using forecasts for them, respectively, we consider these two items independently in the regression analysis (see section 4).

The factor analysis on the stated importance of different forecast parameters also revealed two factors (Table 2; Fig. 2). The variables with highest loadings on factor I1 (α = 0.88) include four different measures of temperature forecast information as well as wind speed and direction, cloudiness, and humidity. Although one might expect all the temperature items to load onto a separate factor, our results suggest that these items grouped together because wind speed and direction, cloudiness, and humidity all influence the effective temperature2 that a person feels (Glickman 2010; OFCM 2003). Thus, we labeled this factor “Importance of temperature-related parameters.” The six variables that primarily loaded onto factor I2 (α = 0.90) represent various aspects of precipitation forecast information, so we labeled it “Importance of precipitation parameters.”

Table 2.

Results of factor analysis on importance of weather forecast parameters. Factor loadings are shown for the parameters that load onto a factor (N = 1461).

Results of factor analysis on importance of weather forecast parameters. Factor loadings are shown for the parameters that load onto a factor (N = 1461).
Results of factor analysis on importance of weather forecast parameters. Factor loadings are shown for the parameters that load onto a factor (N = 1461).
Fig. 2.

Factor analysis on importance of forecast parameters.

Fig. 2.

Factor analysis on importance of forecast parameters.

The separation of temperature and precipitation forecast information into two factors suggests that people perceive the importance of these two types of forecast information differently. Building on findings in LMD09 that, overall, temperature and precipitation were the most important forecast parameters to respondents, these results suggest that effective temperature and precipitation may be most important. A small portion of people rated wind, cloudiness, or humidity forecasts as extremely important, likely because they have specific uses for this information (LMD09). For many, however, interest in forecasts of these parameters appears to be tied primarily to their interest in temperature forecasts. This is corroborated by people’s interest in weather information such as heat index, which combines (hot) temperature and humidity, and wind chill, which combines (cold) temperature and wind speed, and that these combined indices are more indicative of impacts on human health than temperature alone.

We retained the factor scores3 from the factor analyses and used them as variables in the regression analyses discussed in the next section.

4. Regression analysis of sources, uses, and perceptions of forecast information

Next, we conducted linear regression analysis to further examine variability in sources, uses, and perceptions of forecast information. In all of the regressions, we evaluated variance inflation factors (VIFs) using the rule of thumb that a VIF greater than 4.0 would indicate multicollinearity (i.e., high correlation among independent variables). VIFs for all the independent variables in all the linear regressions are less than 3.0 (not reported in the associated tables), indicating that multicollinearity is not an issue (Garson 2009b). For the regressions, we report standardized (beta) coefficients to evaluate the relative effect of each independent variable on the dependent variable.

Seven regressions are presented—one on people’s sources of forecasts (section 4a), four on people’s uses of weather forecasts (section 4b), and two on people’s perceptions of weather forecasts (section 4c). Figure 3 offers a conceptual representation of all the dependent and independent variables used in the regression analyses. Table 3 summarizes the descriptive statistics of all the data used in the regression analyses, described in section 2. The variables are shown grouped into five categories: 1) people’s sources, uses, and perceptions of weather forecast information, 2) people’s weather-related exposure, 3) forecast accuracy, 4) weather variability, and 5) sociodemographics. The same ordering of variables is used to present the regression results.

Fig. 3.

Conceptual representation of linear regression analysis showing all the dependent and independent variables.

Fig. 3.

Conceptual representation of linear regression analysis showing all the dependent and independent variables.

Table 3.

Descriptive statistics of all predictors used in the regression analyses (N = 1461).

Descriptive statistics of all predictors used in the regression analyses (N = 1461).
Descriptive statistics of all predictors used in the regression analyses (N = 1461).

The regression analyses explore a large number of relationships, allowing us to compare the relative influence of a number of independent variables at the same time. We focus our discussion on the results that are most significant, represent the strongest relationships, and indicate conceptual patterns that appear most interesting for improving our understanding of how people obtain, use, and perceive forecast information. Adjusted R2 ranged from 0.10 to 0.23.

a. Influences on people’s frequency of obtaining weather forecast information across sources

Obtaining weather forecasts is a prerequisite for forming perceptions about them and using them. Thus, understanding how and why people get weather forecasts is an important starting point for examining their weather-related attitudes and behaviors. In this section, we investigate what variables influence people’s frequency of obtaining forecasts by performing a regression with the dependent variable as respondents’ total monthly frequency of obtaining forecasts across all sources combined. In other words, we explore what influences how often people obtain forecasts overall, without trying to explain choices among sources in this initial analysis. Results are shown in Table 4.

Table 4.

Linear regression results on total frequency of obtaining forecasts from all sources. Standardized regression coefficients (β) are shown for significant variables (Superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).

Linear regression results on total frequency of obtaining forecasts from all sources. Standardized regression coefficients (β) are shown for significant variables (Superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).
Linear regression results on total frequency of obtaining forecasts from all sources. Standardized regression coefficients (β) are shown for significant variables (Superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).

People’s uses and perceptions of weather forecast information are highly important influences on how frequently they get forecasts. All the forecast “use” variables are significant, indicating that diverse motivations drive how often people obtain forecasts. Among the forecast uses, planning leisure activities (factor U1) is especially influential (β = 0.15, p < 0.01). This may be because people typically have more discretion over leisure activities than the other activities examined in the survey, leading people to seek forecasts more for leisure activity decision making.

All the forecast “perception” variables, except one, are also significant. Temperature-related parameters emerge as being especially influential (β = 0.14, p < 0.01) in people’s frequency of getting forecasts, reflecting the importance of this type of forecast information. We do not have an explanation for why this might be, except that perhaps temperature is relevant for decision making on many days in most parts of the country, whereas precipitation occurs more intermittently and thus precipitation forecasts may not be as relevant to people overall. Understanding the reason underlying this result is an interesting avenue for more in-depth study in future work.

Forecast confidence (β = 0.07, p < 0.01) and importance of NWS information (β = 0.10, p < 0.01) are also significant predictors. This suggests that respondents who have more positive perceptions about weather forecasts tend to get them more often. This may be an iterative relationship such that the more forecasts one gets, the more confidence one has in them. One would expect this relationship to be mediated by how useful the individual found the forecast to be in meeting his/her needs. Interestingly, however, forecast satisfaction is not a significant predictor, indicating that relationships among individuals’ perceptions of forecasts and how frequently they get them may be more complex than one might initially anticipate.

We obtained forecast accuracy and weather variability measures and used them in our regressions because we expected that they would be important influences. A priori, one might expect that people in areas with more variable weather would obtain forecasts more frequently to monitor the changes in weather, and that people in areas with more accurate forecasts would obtain them more frequently because more accurate forecasts would be more useful. Our regression results did not show this. Precipitation forecast accuracy is a statistically significant predictor (β = 0.08, p < 0.01), but greater error in PoP forecasts is associated with individuals obtaining more rather than fewer forecasts. LMD09 showed that chance of precipitation is among the most important forecast parameters to people. The current results suggest that in areas with less accurate PoP forecasts, respondents may be getting forecasts more often to monitor this important information.

The regression results also indicate that a few of the sociodemographic characteristics influence how often people obtain forecasts. In particular, older respondents tend to get forecasts more frequently. This relationship could be due to generational differences in media use and interests in weather and weather forecasts. People who have resided near their current location longer also tend to get forecasts more frequently, perhaps because they are more familiar with the forecasts and forecast sources (such as local television broadcasters) once they have lived in an area longer. Future work would be needed explore these relationships and potential explanations in more detail.

Overall, this regression indicates that key drivers explaining the frequency with which people obtain forecasts are what they use forecasts for and their perceived importance of and confidence in forecast information. People’s experience with weather forecasts, as measured by forecast accuracy, and sociodemographic characteristics also influence frequency of obtaining forecasts, but to a lesser extent.

b. Influences on people’s uses of weather forecast information for activities

To explore what influences people’s uses of forecasts for different activities, we present results from four regressions on use measures (Table 5): forecast use for planning leisure activities (factor U1), for planning work/school-related activities (factor U2), for planning how to dress yourself/children, and for simply knowing what the weather will be like.

Table 5.

Linear regression results on the uses of weather forecasts examined in the study. Standardized regression coefficients (β) are shown for significant variables (superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).

Linear regression results on the uses of weather forecasts examined in the study. Standardized regression coefficients (β) are shown for significant variables (superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).
Linear regression results on the uses of weather forecasts examined in the study. Standardized regression coefficients (β) are shown for significant variables (superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).

Frequency of getting forecasts is a significant, positive predictor in all four regressions, suggesting that respondents who get forecasts more often use them more for planning all of the activity types examined. Temperature-related and precipitation parameters are also significant positive predictors in all four regressions, indicating that temperature-related and precipitation forecast information is relevant to multiple decision-making contexts. These results, combined with those in section 4a, offer empirical support for the positive relationships among frequency of obtaining forecasts, importance of forecast parameters, and use of forecasts that one might expect.

One would also expect that positive perceptions of forecasts are interrelated with forecast use for different activities. Interestingly, however, few of the perception variables tested were significant influences in this analysis. As in section 4a, forecast satisfaction was not a significant predictor. Forecast confidence was only a weak predictor of forecast use for leisure activities.

Regarding people’s experiences with weather, the more leisure time they spent outdoors, the more respondents tended to use forecasts for planning leisure activities (β = 0.18, p < 0.01). Similarly, the more work time they spent outdoors, the more they used forecasts for planning work/school-related activities (β = 0.29, p < 0.01). Although these relationships may be anticipated, these results empirically show that respondents’ weather-related exposure in different domains of their lives is important because it influences how they use forecasts for planning activities in those domains. Individuals who live in areas with increased variability in precipitation and maximum temperature use forecasts more for leisure activities (β = 0.06, p < 0.05) and planning how to dress (β = 0.13, p < 0.01), respectively. This suggests that some people who live in areas where the weather is more variable may respond to that variability by increasing their use of forecasts to plan for certain activities.

Among the significant sociodemographic variables, full-time employees use forecast information more frequently for work/school-related purposes than non-full-time employees (β = 0.13, p < 0.01), as one might expect. Females tend to use forecasts more than males, particularly for planning how to dress (r = −0.17, p < 0.01), but also for simply knowing about the weather (r = −0.09, p < 0.01) and for planning leisure activities (r = −0.07, p < 0.01). These findings suggest that, on average, there are gender differences in how people use forecast information. Possible reasons for this may include gender differences (on average) in attitudes and behaviors toward dressing and in family roles (e.g., Amato et al. 2003; Craig 2006).

Overall, these four regressions show some themes underlying respondents’ use of forecasts for different types of activities. The variables showing the strongest, most consistent relationships across forecast uses are the frequency of obtaining forecasts and the importance of temperature-related and precipitation forecast information. Other variables, such as individuals’ leisure or work-related weather exposure, their experience with weather, and their sociodemographic characteristics, are related to forecast uses for certain types of activities.

c. Influences on people’s stated importance of weather forecast parameters

To begin exploring influences on people’s perceptions of forecast information, we performed regressions on people’s stated importance of temperature-related (factor I1) and precipitation (factor I2) forecast parameters (Table 6).

Table 6.

Linear regression results on the importance of temperature-related and precipitation forecast parameters. Standardized regression coefficients (β) are shown for significant variables (superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).

Linear regression results on the importance of temperature-related and precipitation forecast parameters. Standardized regression coefficients (β) are shown for significant variables (superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).
Linear regression results on the importance of temperature-related and precipitation forecast parameters. Standardized regression coefficients (β) are shown for significant variables (superscript letters a, b, and c indicate significance at the 10%, 5%, and 1% level, respectively; N = 1461).

Frequency of obtaining forecasts was a significant predictor of the importance of temperature-related information but not precipitation information. This is related to the results in section 4a, in which the importance of temperature-related information was a stronger predictor of frequency of obtaining forecasts. Forecast uses are significant, positive predictors of forecast parameter importance in six of the eight possible relationships in the regressions. This suggests that, as discussed in section 4a, people’s uses of forecasts are interrelated with how important they deem temperature and precipitation information to be. Compared to the other use variables, use of forecasts for planning leisure activities was the most influential on respondents’ stated importance of temperature-related (β = 0.23, p < 0.01) and precipitation (β = 0.10, p < 0.01) parameters. This again suggests that activities for which individuals have more discretion in decision-making (e.g., leisure vs work) may have more influence on the importance of forecast information. These results, when considered with the results from the previous regressions on sources and uses, further support the notion that individuals’ uses of forecasts, positive perceptions of forecasts, and frequency of obtaining forecasts are all dynamically interrelated. We cannot identify directions of influence in the current analysis, but we believe the relationships may be multidirectional or iterative.

Confidence in forecasts and the perceived importance of NWS information were also significant, positive predictors of the importance of temperature and precipitation-related parameters. Although these questions were asked in different ways in different parts of the survey, the relationships among the importance variables corroborate each other and reveal consistency in people’s positive perceptions about weather forecast information. As in previous regressions, however, satisfaction with forecasts was not a significant predictor.

Overall, there were few significant relationships between the importance variables and the variables representing people’s weather-related exposure and their experiences with weather and weather information. We expected that these variables, especially weather variability, would be a stronger influence on the importance of forecast parameters, but the analysis did not strongly support this. The analysis does show a weak relationship between increased precipitation variability and increased importance of precipitation parameters (β = 0.05, p < 0.10). This suggests that, related to the discussion in section 4a about precipitation forecast error, respondents in areas with greater variability in precipitation may find precipitation forecast information more important because they need to more closely monitor precipitation information.

Many of the sociodemographic characteristics—employment, race, age, education, years of residence, and income—are statistically significant influences on the importance of temperature and precipitation parameters. It is interesting to note that the standardized coefficients are all positive for the temperature-related dependent variable and are all negative for the precipitation-related dependent variable. We currently do not have an explanation for these relationships, although it appears that people with characteristics associated with lower socioeconomic status (e.g., less educated, lower income) tend to find temperature-related information more important, while those with higher socioeconomic status tend to find precipitation information important.4 Additional research could further explore these relationships, including whether these respective set of characteristics are associated with subpopulations that possess distinct perceptions of forecast parameters, or whether there are other mediating variables.

Overall, results from these two regressions again show the interrelationships among frequency of obtaining forecasts, forecast use, and importance of forecast parameters. They also indicate that confidence in forecasts and importance of forecast parameters are related, and that certain sociodemographic characteristics appear to be associated with different perceptions of the importance of temperature-related versus precipitation forecast parameters.

5. Summary and discussion

Providing usable information is an important goal of the weather enterprise, and a key part of achieving this goal is understanding how people obtain, use, and perceive weather forecast information and why. To address this complex and inherently multidisciplinary area, studies that draw on a range of concepts and approaches are needed. In this paper, we explore 1) patterns in individuals’ sources, uses, and perceptions of weather forecasts using factor analysis; and 2) relationships among individuals’ sources, uses, and perceptions of weather forecasts, sociodemographic characteristics, weather-related exposure, and experiences with forecast accuracy and weather variability using regression analysis.

The factor analysis on the sources of forecasts did not reveal any factors that met all criteria, suggesting a lack of clear patterns in respondents’ behaviors regarding the types of sources they access and the frequency with which they do so. This may be due to the limited use of certain sources among our survey respondents. Technology is rapidly changing the sources from which people can get weather forecasts (Rockwell et al. 2007), with information now available via such channels as smart phones, social media platforms, and 24/7 weather alert notifications. It is unclear how these new sources will change the ways people get weather forecast information. The additional sources may further increase the variance in people’s forecast-acquisition behaviors, or concrete patterns of source usage may begin to emerge.

People’s uses and perceptions of forecast information are broad, multifaceted concepts. We examined patterns in one dimension of each concept by performing factor analysis on forecast uses for different activities and stated importance of different forecast parameters. The analysis revealed that forecast use for planning leisure activities and work/school-related activities constituted different factors, as did the importance of temperature-related parameters and precipitation parameters. These results indicate differences in how different people use forecast information and in how they think about forecasts. More generally, this analysis affirms that empirical data from members of the public can be used to identify and characterize underlying heterogeneity in the use and perceptions of forecasts, revealing more meaningful information about people’s attitudes and behaviors than what aggregate level descriptive data can show.

Regression analyses supported some things we expected intuitively, but they also identified some new relationships and failed to reveal some relationships we anticipated. Rather than make a priori assumptions about the directions of relationships, we used the measures of people’s frequency of obtaining forecasts, forecast uses for different activities, and stated importance of forecast parameters as both dependent and independent variables. The results show that there are interrelationships among these three sets of measures, suggesting that how people obtain, use, and perceive forecasts all positively and strongly influence each other. To assess causal relationships in greater depth, future work is needed. Such understanding about why people do what they do regarding weather forecasts could then be used to help develop weather information messages that better address audiences’ specific motivations, constraints, capabilities, and needs. Building messages based on such understanding would greatly advance weather community efforts to communicate more useful information.

Of the other forecast perception measures that were included as independent variables, forecast confidence and importance of NWS information were significant in many of the regressions, particularly in predicting total frequency of obtaining forecasts and importance of forecast parameters. However, forecast satisfaction was not significant in any of the regressions. This could be due to the way we measured satisfaction; a single-item measure may not fully capture the complexity of the concept. Further work is needed to understand what forecast satisfaction means to individuals and how it relates to their attitudes and behaviors regarding forecast information. More generally, work is needed to explore how people perceive concepts such as confidence, importance, and satisfaction with respect to weather forecast information, as our results suggest these are different constructs. It is also important to better understand how people perceive these concepts to the extent they are used as metrics of the quality or value of forecast services. Ultimately these and related constructs should be investigated in relation to behavioral responses as a basic goal of the provision of forecasting is to inform decision making.

The regression results also revealed that respondents’ work (leisure)-related exposure to weather was positively related to their use of forecasts for planning work (leisure) activities. These findings indicate that the extent to which people are exposed to weather in different aspects of their lives is related in a consistent manner to their forecast behaviors. Several different sociodemographic characteristics were significant in some of the regressions, indicating that there are differences in how people think about and behave toward weather forecast information given their age, income, gender, ethnicity, and other characteristics. Building on this initial knowledge may be useful for developing messages to meet the needs of different and diverse segments of the population.

We also included measures of weather variability and forecast accuracy as independent variables in the regressions to represent people’s experiences with weather and weather forecasts, but these measures were significant in only some cases. This may indicate that different measures of variability and accuracy need to be tested, or that people’s weather-related experiences influence their attitudes and behaviors in more complex ways than reflected by these measures alone. It may also indicate that people’s perceptions of forecast accuracy and variability differ from actual measures of accuracy and variability. Supporting this idea, there was a weak correlation between individuals’ confidence in 1-day forecasts and precipitation variability (r = 0.061, p < 0.05), but no significant correlation between forecast confidence and temperature forecast accuracy, precipitation forecast accuracy, or temperature variability.

This study is a first effort at elucidating these types of patterns and relationships. Future work could design and test additional survey items and analyze the resulting data to examine the robustness of the factors we found, explore whether other dimensions of forecast uses and forecast perceptions emerge, and tie these to explanatory behavioral theories. Additional dimensions of forecast use, for example, may include forecast use for daily, regular activities (e.g., dressing for the day), information surveillance (e.g., for severe weather), or entertainment. Additional dimensions of perceptions may include components of forecast confidence, trust, or credibility. To complement a deeper understanding of uses and perceptions, one could also investigate patterns in other broad concepts, such as forecast value or forecast interpretation.

In summary, this study uses concepts and tools from several disciplines to begin to characterize and explain patterns in people’s sources, uses, and perceptions of weather forecasts. Here, we used a survey and quantitative analysis to explore these issues, but complementary efforts using other approaches are also needed. Further studies can build on additional social science concepts and theories and employ other research methods (such as in-depth interviews or behavioral experiments). Triangulation of results using different theoretical approaches and methods can both support the validity of findings and help develop a deeper understanding of individuals’ attitudes and behaviors with respect to weather forecasts. Ultimately this understanding is necessary to help forecast developers and providers make informed decisions about improving the content, format, and channels of weather forecast information to better meet users’ needs.

Acknowledgments

Thank you to Brent MacAloney for his assistance with the NWS verification data, Sheldon Drobot for his assistance with the NCDC observation data, and Barbara Brown for her input about the verification metrics. Thank you also to Douglas Hilderbrand, Jason Knievel, and five anonymous reviewers for their constructive comments on the manuscript. This work is supported by NCAR’s Collaborative Program on the Societal Impacts and Economic Benefits of Weather Information (SIP), which is funded by the National Science Foundation and the National Oceanic and Atmospheric Administration through the U.S. Weather Research Program. NCAR is sponsored by the National Science Foundation. The views and opinions in this paper are those of the authors.

APPENDIX

Survey Questions

The survey questions discussed in the manuscript, except for the sociodemographic questions, are presented below using the question numbers and order from the full survey. Subquestions are denoted by roman numerals. All questions were asked of all N = 1461 respondents. The question wording is reproduced below, but the formatting (spacing, typesetting, etc.) has been altered for space considerations.

Q2. How often do you get weather forecasts from the sources listed below? (Order of subquestions randomized. Response choices for each subquestion: Rarely or never, Once or more a month, Once a week, Two or more times a week, Once a day, Two or more times a day)

  • Local TV stations

  • Cable TV stations (e.g., CNN, The Weather Channel)

  • Newspapers

  • Telephone (dial-in) weather information source

  • Commercial or public radio

  • NOAA Weather Radio

  • National Weather Service (NWS) webpages

  • Other webpages

  • Cell phone, personal desk assistant (PDA), pager, or other electronic device

  • Friends, family, coworkers, etc.

Q5. On average, year round, how often do you use weather forecasts for the activities listed below? (Order of subquestions randomized. Response choices for each subquestion: Rarely or never, Less than half of the time, About half of the time, More than half the time, Usually or always, Not applicable to me)

  • Planning how to dress yourself or your children

  • Planning how to get to work or school

  • Planning to do yard work or outdoor house work

  • Planning job activities

  • Planning social activities

  • Planning travel

  • Planning weekend activities

  • Simply knowing what the weather will be like

Q6. A weather forecast can provide several types of information about temperature, cloudiness, winds, and precipitation (such as rain, snow, hail, or sleet). How important is it to you to have the information listed below as part of a weather forecast? (Order of subquestions randomized. Response choices for each subquestion: Not at all important, A little important, Somewhat important, Very important, Extremely important)

  • Chance of precipitation

  • Amount of precipitation

  • Type of precipitation

  • When precipitation will occur

  • Where precipitation will occur

  • Chance of different amounts of precipitation (e.g., greater than ½ inch, 1 inch, 6 inches)

  • Low temperature

  • High temperature

  • What time of day the high temperature will occur

  • What time of day the low temperature will occur

  • How cloudy it will be

  • Wind speed

  • Wind direction

  • Humidity levels

Q7. Overall, to what extent are you satisfied or dissatisfied with the weather forecast information that you currently receive? (Response choices: Very dissatisfied, Dissatisfied, Neither satisfied nor dissatisfied, Satisfied, Very satisfied)

Q11. Weather forecasts are available for up to 14 days into the future. This means that a 1-day forecast is for the weather 1 day (24 h) from now, that a 2-day forecast is for the weather 2 days (48 h) from now, and so on. How much confidence do you have in weather forecasts for the times listed below? (Response choices for each lead time: Very low, Low, Medium, High, Very high)

Forecasts for weather …

  • Less than 1 day from now

  • 1 day from now

  • 2 days from now

  • 3 days from now

  • 5 days from now

  • 7 to 14 days from now

Q23. On average, year round, what percent of your on-the-job time (i.e., the time that you are paid for working) is spent outdoors? [Response choices: 0% (none of my time), 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100% (all of my time), not applicable to me]

Q24. On average, year round, how many hours per week do you spend traveling outside to and from work or school in a mode that could be affected by the weather? (Response choices: open-ended, not applicable to me)

Q25. On average, year round, what percent of your leisure time is spent outdoors? [Response choices: 0% (none of my time), 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100% (all of my time), not applicable to me]

Q26. On average, year round, how many hours per week do you spend working outside in your yard or garden, washing your car, working on the house, or doing other outdoor activities around the house? (Response choices: open-ended).

REFERENCES

REFERENCES
Amato
,
P.
,
D.
Johnson
,
A.
Booth
, and
S.
Rogers
,
2003
:
Continuity and change in marital quality between 1980 and 2000
.
J. Marriage Fam.
,
65
,
1
22
.
CFI Group
,
2005
:
National Weather Service customer satisfaction survey: General public
.
Report to the National Oceanic and Atmospheric Administration, 154 pp. [Available online at http://www.weather.gov/com/files/NWS_Public_survey050608.pdf.]
Craig
,
L.
,
2006
:
Does father care mean fathers share? A comparison of how mothers spend time with children
.
Gender Soc.
,
20
,
259
281
.
Dillman
,
D. A.
,
2000
:
Mail and Internet Surveys: The Tailored Design Method
.
2nd ed. John Wiley & Sons, 464 pp
.
Garson
,
G. D.
, cited
2009a
:
Factor analysis
.
Statnotes: Topics in Multivariate Analysis. [Available online at http://faculty.chass.ncsu.edu/garson/PA765/factor.htm.]
Garson
,
G. D.
, cited
2009b
:
Multiple regression
.
Statnotes: Topics in Multivariate Analysis. [Available online at http://faculty.chass.ncsu.edu/garson/PA765/regress.htm.]
Glickman
,
T.
, Ed.,
2000
:
Glossary of Meteorology
.
2nd ed. Amer. Meteor. Soc., 855 pp. [Available online at http://amsglossary.allenpress.com/glossary
.]
Harris Poll
,
2007
:
Local television news is the place for weather forecasts for a plurality of Americans
.
The Harris Poll #118, issued November 28, 2007. [Available online at http://www.harrisinteractive.com/vault/Harris-Interactive-Poll-Research-Weather-2007-11.pdf.]
Hatcher
,
L.
,
2007
:
A Step by Step Approach to Using SAS for Factor Analysis and Structural Equation Modeling
.
9th ed. SAS Institute, 588 pp
.
Karl
,
T. R.
,
R. W.
Knight
, and
N.
Plummer
,
1995
:
Trends in high-frequency climate variability in the twentieth century
.
Nature
,
377
,
217
220
.
Kaye
,
B.
, and
T.
Johnson
,
2002
:
Online and in the know: Uses and gratifications of the web for political information
.
J. Broadcast. Electron. Media
,
46
,
54
71
.
Lasswell
,
H. D.
,
1948
:
The structure and function of communication in society
.
The Communication of Ideas, L. Bryson, Ed., Harper, 37–51
.
Lazo
,
J. K.
, and
L. G.
Chestnut
,
2002
:
Economic value of current and improved weather forecasts in the U.S. household sector
.
Report to the NOAA Office of Policy and Planning, 213 pp. [Available online at http://www.economics.noaa.gov/bibliography/economic-value-of-wx-forecasts.pdf
.]
Lazo
,
J. K.
,
R. E.
Morss
, and
J. L.
Demuth
,
2009
:
300 billion served: Sources, perceptions, uses, and values of weather forecasts
.
Bull. Amer. Meteor. Soc.
,
90
,
785
798
.
Morss
,
R. E.
,
J. L.
Demuth
, and
J. K.
Lazo
,
2008a
:
Communicating uncertainty in weather forecasts: A survey of the U.S. public
.
Wea. Forecasting
,
23
,
974
991
.
Morss
,
R. E.
,
J. K.
Lazo
,
B. G.
Brown
,
H. E.
Brooks
,
P. T.
Ganderton
, and
B. N.
Mills
,
2008b
:
Societal and economic research and applications for weather forecasts: Priorities for the North American THORPEX program
.
Bull. Amer. Meteor. Soc.
,
89
,
335
346
.
Morss
,
R. E.
,
J. K.
Lazo
, and
J. L.
Demuth
,
2010
:
Examining the use of weather forecasts in decision scenarios: Results from a U.S. survey with implications for uncertainty communication
.
Meteor. Appl.
,
17
,
149
162
.
NRC
,
2006
:
Completing the Forecast: Characterizing and Communicating Uncertainty for Better Decisions Using Weather and Climate Forecasts
.
National Research Council, National Academies Press, 124 pp
.
NRC
,
2010
:
When Weather Matters: Science and Service to Meet Critical Societal Needs
.
National Research Council, National Academies Press, 159 pp
.
NWS
,
2009
:
Verification procedures: National Weather Service instruction 10-1601
. .]
OFCM
,
2003
:
Report on wind chill temperature and extreme heat indices: Evaluation and improvement projects
.
Office of the Federal Coordinator for Meteorological Services and Support Research Rep. FCM-R19-2003, 75 pp
.
Papacharissi
,
Z.
, and
A. M.
Rubin
,
2000
:
Predictors of internet use
.
J. Broadcast. Electron. Media
,
44
,
175
196
.
Papacharissi
,
Z.
, and
A. L.
Mendelson
,
2007
:
An exploratory study of reality appeal: Uses and gratifications of reality TV shows
.
J. Broadcast. Electron. Media
,
51
,
355
370
.
Pew Research Center
,
2008
:
Key news audiences now blend online and traditional sources: Audience segments in a changing news environment
.
Pew Research Center Biennial News Consumption Survey, 129 pp. [Available online at http://people-press.org/files/legacy-pdf/444.pdf
.]
Presser
,
S.
,
2004
.
Questions and Answers in Attitude Surveys
.
Sage Publications, 388 pp
.
Rockwell
,
M.
,
G.
Dickson
, and
P. J.
Bednarski
,
2007
:
Cold, hard facts straight from the cellphone: Consumers begin turning to wireless devices for forecasts
.
Broadcasting Cable
,
137
,
20
.
Tan
,
A. K. O.
,
1976
:
Public media use and preference for obtaining weather information
.
Journalism Q.
,
53
,
694
705
.
Toth
,
Z.
,
O.
Talagrand
,
G.
Candille
, and
Y.
Zhu
,
2003
:
Probability and ensemble forecasts
.
Forecast Verification: A Practitioner’s Guide in the Atmospheric Sciences, I. T. Jolliffe and D. B. Stephenson, Eds., John Wiley and Sons, 137–163
.
Tourangeau
,
R.
,
L. J.
Rips
, and
K.
Rasinski
,
2000
:
The Psychology of Survey Response
.
Cambridge University Press, 415 pp
.
U.S. Census
, cited
2007
:
2006 American community survey
. .]

Footnotes

1

All statistical data analysis was conducted using SPSS 17.0.

2

The American Meteorological Society defines effective temperature as the temperature at which motionless saturated air would induce, in a sedentary worker wearing ordinary indoor clothing, the same sensation of comfort as that induced by the actual conditions of temperature, humidity, and air movement.

3

Factor scores were retained using the Anderson–Rubin method. The scores have a mean of zero and a standard deviation of one, and they are uncorrelated between factors within the same analysis.

4

We thank an anonymous reviewer for this observation.