:SITY OF ANA-CHAMPAIGN BOOKSTACKS Digitized by tine Internet Archive in 2011 with funding from University of Illinois Urbana-Champaign http://www.archive.org/details/ridesharingtowor345horo Faculty Working Papers RIDESHARING TO WORK: A PSYCHOSOCIAL AivTALYSIS Abraham D. Horowitz and Jagdish N. Sheth #345 College of Commerce and Business Administration University of Illinois at Urbana-Champaign FACULTY "WORKING PAPERS College of Connnerce and Business Administration University of Illinois at Urbana-Champaign October 13, 1976 RIDESHARING TO WORK: A PSYCHOSOCIAL ANALYSIS Abraham D. Horowitz and Jagdish N. Sheth #345 It, ■ ii-j.-.':^- J, '; }."' ' ' RIDESHARING TO WORK: A PSYCHOSOCIAL ANALYSIS by Abrahaip D. Horowitz Transportation and Irban Analysis Department Research Laboratories General Motors Corporation Warren, Michigan 48090 Professor Jagdish N. Sheth Department of Btisiness Administration University of Illinois Urbana, Illinois 61801 ABSTRACT A mathematical model of ridesharing (carpooling) was proposed and tested using data collected for Research Laboratories, General Motors Corporation, by an independent marketing research firm. Developed from factor analysis and analysis of variance, the model postulates that emotional and behavioral predispositions toward ridesharing to work are determined in an additive-interactive manner by the perceived relative advantages and disadvantages of ridesharing. Two latent cognitive factors underlying the perception of ridesharing emerged from the analysis: Time-Convenience (ridesharing disadvantages) and Private-and-Pub lie-Cost (ridesharing advantages). Analyses of the attitudes of commuters who drive alone and who share a ride to work •»^^■'■(rI.ate the hypotheses of the attitudinal model: (1) The emotional (affective) and the behavioral (intentional) predispositions toward ridesharing are additively determined by the two cognitive factors, but the Time- Convenience factor is a markedly be. ter explanatory variable than the Private-and-Public-Cost factor. (2) The interactive term inqjlies that solo drivers with strong perceived ridesharing disadvantages do not like ridesharing regardless of the magnitude of the perceived advantages; no interaction was found for carpoolers. A segmentation technique based on the two cognitive factors was developed and applied to solo drivers' socioeconomic characteristics; the results show that males with a relative high socioeconomic status are more sensitive to the Private-and-Public-Cost factor than are other Individuals. The results supply diagnostic information which suggests how to find promotional methods to improve the image and use of ridesharing. TABLE OF CONTENTS Page ABSTRACT i INTRODDCTION 1 Background ; 1 Data. 3 Nomenclature 4 Questionnaire 4 RESULTS 7 Demographic and Travel Characteristics 7 Ridesharing Cognitive Profile .' 14 Structuring Cognition 17 Drive-Alone Cognitive Profile 20 Differences among Ridesharing and Drive-Alone Cognitive Profiles . 22 Affect toward Ridesharing and the Intention to Share a Ride. ... 26 Models Relating Cognition Factors to Affect and Intention 26 Test of the Affect Model 30 Test of the Intention Model 33 A Marketing Segmentation Technique 34 DISCUSSION , 37 Future Research 38 ACKNOWLEDGEMENTS 39 REFERENCES 40 APPENDICES 43 INTRODUCTION Background Consider a future situation in which public officials desire to increase the level of ridesharing. What are the advantages and disadvantages of ridesharing? How are these advantages and disadvantages perceived by people generally? How are they perceived by people potentially amenable to ridesharing? How do these people use this information in making their travel decisions? How might the propensity toward ridesharing be enhanced? This study alms to answer these questions by • investigating the ridesharing attitude structure of individuals and by identifying homogeneous subgroups who differ in their • attitudes. The literature on ridesharing, developed mainly as a consequence of the energy shortage of 1973-4, is concerned with the travel characteristics of carpoolers (Herman and Lam, 1975), with ridesharing matching (Berry, 1975; Davis et al., 1975; Kendall, 1975; Rosenbloom and Shelton, 1974) , with the study of incentives for inducing people to share a ride (Ben-Akiva and Atherthon, 1976; Margolin and Misch, 1976; Zerega and Ross, 19/6), and with clinical-social aspects (Blankenship, 1975; Barkow, 1976) . Studies on matching and ridesharing incentives are based on the presumption that solo drivers can be induced to carpool by offering them direct incentives (for example, parking and traffic priorities) or that driving alone might be discouraged by, say, raising the cost of gasoline. Effective promotion of ridesharing requires a direct knowledge of how it is viewed both by commuters who drive alone and by those who share a ride to work. Attitudes toward ridesharing have been investigated by Alan M. Voorhees and Associates (1973), by Carnegie Mellon University (1975), and by Dueker and Lewin (1976) . The Alan M. Voorhees and Carnegie Mellon studies showed that there are profound differences in attitudes toward rldesharing between solo drivers and carpoolers. However, they did not study the structure of attitudes in depth, nor did they attempt to identify homogeneous subgroups that differ in their attitudes. Dueker and Lewin report ongoing research at the University of Iowa employing laboratory studies and models in experimental psychology, but the results have not been published at the time of this writing. Horowitz (1975) developed a theoretical framework for the measure- ment of attitudes toward ridesharing and driving alone and presented mathematical models relating mode choice to the perceived advantages and disadvantages of ridesharing and to other attitudinal and socioeconomic characteristics. The model is an adaption of the Howatd-Sheth (1974) model of attitude-behavior relationship to the area of carpooling behavior. For the purpose of testing this framework and achieving the aims of this research, the Research Laboratories contracted with a marketing research firm to collect the required data. Data A survey was conducted among residents of the Chicago metropolitan area contacted through their employers. The main reason for choosing Chicago as the site of the data collection was that it offers a wide variety of businesses both in terms of type and size in both the city and its suburbs and a variety of public transit services. Personnel departments of 43 firms, chosen randomly from a large list of companies that employ at least 100 people, were first contacted. Table 1 summarizes the distribution by size and location of employers who expressed their willingness to participate. About 60% of these fiirms are manufacturing companies, while the others are distributors. EMPLOYERS DISTRIBUTION BY SIZE AND LOCATION NUMBER OF EMPLOYEES PER FIRM CHICAGO SUBURBS TOTAL 100 - 300 301 - 1 500 1 501 - 8 000 2 6 7 9 6 4 11 12 11 15 19 34 Table 1 insurance companies, and other types of organizations. Personnel departments were asked to contact roughly equal numbers of carpoolers, solo drivers, and public transit users to answer a self administered mail back type questionnaire but hand delivered. During the fall and winter of 1975 two thousand questionnaires were distributed of \*hich 1020 questionnaires were returned. After eliminating questionnaires with a large amount of missing data, 822 questionnaires remained for analysis: 323 carpoolers, 382 solo drivers, and 117 public transit commuters. Since in this sample virtually all carpoolers owned at least one car while 75% of transit users did not own cars it was assumed that car ownership is a necessary condition for sharing a ride to work. It was decided, then, to analyze data relating to carpoolers and solo drivers only. The method of contacting commuters through their employers (a method seldom used in transportation research) proved to have certain advantages over the traditional methods of data collection. First, the rate of return was relatively high (about 50%) compared to mail surveys. Second, the cost for data collection was smaller than that required for home-interviews. Nomenclature Throughout this paper the two underlying modes of travel will be called "drive-alone" and "ridesharing," and the two types of commuters "solo drivers" and "carpoolers," respectively. The concept of ridesharing is restricted in the present study to the use of privately owned cars. Questionnaire Three types of information were collected through the questionnaire: the first two are socioeconomic and travel characteristics. The third type is attitudinal data with respect to both ridesharing and driving alone. A few words are in order to describe the theoretical approach that guided the formulation of the attitudinal questions. There is a consensus among attitude researchers (Rosenberg, 1960; Fishbein, 1967; Sheth, 1974) that attitudes consist of one or more of three elements: (1) cognitive evaluations or beliefs, (2) affect (like-dislike emotional tendency), and (3) behavioral intention. Cognitive Evaluations . It is hypothesized that an individual has a set of evaluative beliefs about ridesharing and drive-alone modes of travel to work with respect to cost, time saving, convenience, etc. Ten such attributes presented in the upper part of Figure 1 were elicited from informal interviews conducted individually with a few carpoolers and solo drivers. The cognitive evaluations are measured on a seven-point scale from "very low" to *'very high." Affect represents the positive or negative emotional predisposition toward an object, and is presumed to be unidimensional although it is possible that there is a complex cognitive structure underlying it. A measure of the affect toward ridesharing was obtained by the use of the rating scale shown in Figure 1. Intention . Ridesharing intention refers to the stated plan of an individual to carpool and was measured by the last question that appears in Figure 1. Intention is hypothesized also to be related to the cognitive profile of evaluations. Intention is a qualified expression of behavior: given a span of time, when behavior is likely to be manifested, the individual estimates at the beginning of the period of time whether he or she would or would not behave in a certain manner. Since the shorter the period of time between intention and behavior, the more valid is the intention (Fishbein, 1967), the time span was limited to the next two or three months. A theoretical structure of the relation between the cognitive evaluations, affect, and intention will be presented later in this paper. EVALUATIONS (COGNITIVE PROFILE) SCALES EXPENSIVE COMFORTABLE PLEASANT RELIABLE SAVES TIME CONVENIENT SAFE FROM CRIME ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION DRIVING ALON. VERY LOW 2 3 4 [] [] [ [] [] [ [] [] [ [] [] [ [] [] [ C3 [] [ [] [] C [] [] [ [],[] [ VERY HIGH 5 6 M CARPOOLING VERY LOW VERY GH 2 3 4 5 6 AFFECT SCALE All things considered, which statement best describes how you like the idea of YOU being a member of a carpool? LIKE EXTREMELY [] DISLIKE SLIGHTLY [] LIKE MODERATELY [] DISLIKE MODERATELY [] LIKE SLIGHTLY [] DISLIKE EXTREMELY [] NEITHER LIKE NOR DISLIKE [] INTENTION SCALE How likely are you to join a carpool within the next two or three months? DEFINITELY WILL VERY LIKELY SOMEWHAT LIKELY [] SOMEWHAT UNLIKELY [] [] VERY UNLIKELY [] [] DEFINITELY WILL NOT [] CANNOT SAY [] Figure RESULTS Demographic and Travel Characteristics Table 2 summarizes the socioeconomic characteristics of the solo driver and the carpooler groups. A MANOVA (multivariate analysis of variance) test using Wilks Lambda criteria (Morrison, 1967) performed on 13 variables showed that solo drivers differ significantly from K carpooler s (F= 5.8, p <_ 0.001). The column "Prob. <^" indicates the probability at and under vfhich the difference between the means of the tV70 groups is due to chance. The variables presented in Table 2 are self-explanatory and are ordered according to their power to discriminate between the two groups. For descriptive purposes a univariate t-test was performed on each variable. 2 The column "F-ratio" displays the value of t . The column "Explained Variance" for the multivariate test is the square of the canonical correlation between the' vector of variables and the artificial MANOVA variables (a vector of O's and I's according to that group to which an observation belongs) expressed in percentages . For a univariate test the explained variance is the percentage of the sum of squares betv/een groups of the total sum of squares. This measure 2 equals, for large samples, the u measure developed by Hays (1963) that has been increasingly used in psychology and consumer research (for example Bettman, et al. 1975). The column "Prob. <^' indicates the level of significance between 2 the means, as it may be between the groups. The o) measure indicates the dlscriminab-flity between the groups that is, the degree of non-overlapping of the two distributions. Consistent v;ith the position of Anderson (1961) , who showed that an interval scale is not a prerequisite to making a statistical Inference based on a parametric test, two of the variables in Table 2 are measured on an ordinal scale: size of the car and professional status. Moreover, the assumption of an interval scale is not critical when the test is based on a large number of observations (Krishnan and Clelland, 1973). u^ SOCIOECONOMIC COMPARISONS AMONG SOLO DRIVERS AND CARPOOLERS SOLO DRIVERS CARPOOLERS F-R.'^TIO** PROB.5 EXPLAINED VARIANCE MEAN^I n SO, MEANg ,L SO2 MULTIVARIATE STATISTICS MANOVA — 382 ^-_ _-. 323 .-- 5.8 0.001 9.8 UNIVARIATE STATISTICS CAR SIZE (l=subcompact, 2=compact, 3=intermediate, 4=fun size) 2.5 328 2.5 2.9 259 1.0 25.5 0.001 ^•1 YEARS AT PRESENT EMPLOYER 8,3 374 9.0 11.2 318 9.5 17.2 0.001 2.4 ilARITAL STATUS (l=s ingle, 2=married) 1.67 381 0.5 1.75 323 0.4 7.2 0.007 1.0 YEARS AT PRESENT RESIDENCE 7.1 375 6.9 8.6 317 9.0 6.2 0.013 0.6 AGE (l=under 25, 2=25-34, 3=35-44, 4=45-54, 5=55- 64, 6=65 or over) 2.74 380 1.3 2.93 322 1.2 . 4.1 0.044 0.6 HOUSEHOLD SIZE 3.0 356 1.4 3.3 316 1.6 3.8 0.05 0.5 NUMBER OF LICENSED DRIVERS IN HOUSEHOLD 2.2 378 1.0 2.2 315 0.9 0.0 :i.s. „. HOUSEHOLD AUTO OWNERSHIP 1.9 379 0.9 1.8 311 0.8 1.3 N.S. — CAR AGE (years) 3.6 342 2.8 3.7 260 2.3 0.4 N.S. — SEX (l=male, 2=female) 1.45 381 0.5 1.50 323 0.5 1.8 N.S. ••• HOUSEHOLD ANNUAL INCOME (2=$3 001-5 000, 3=$5 001- 7 000, etc.. 9=$17 001- 19 0O0,...,13=$25 001- 27 000, 14=$27 OOl and more) 8.9 331 2.8 9.1 278 2.8 1.0 N.S. - — OCCUPATION (l=professional, 2=nianager, 3=clerical worker, 4=craftsman, 5=operator, 6=service worker) 2.46 377 1.1 2.38 319 1.1 0.9 A.S. - — EDUCATION {3=attended high school, 4=graduated high school, 5=attended college, 6=finish9d college) 5.0 377 1.1 5.0 321 1.2 0.0 N.S. — - ♦N-NUMBER OF INDIVIDUALS SD-STANDARD DEVIATION N.S.-;J0T SIGNIFICANT ♦•-DEGREES OF FREEDOli: 13 AND 691 FO.". THE 1 AND :i^+N2- ULTIV.nRIATE TEST 2 FOR EACH UNIVARIATE TEST T-'.ble 2 The socioeconomic variable that discriminates most between the two groups is size of the car owned: carpoolers own larger cars than solo drivers. Note that in spite of the significant difference (p <^ 0.001), the explained variance for this variable is only 4.1%. A better and direct description is provided in Figure 2, that shows that differences in car size ownership are mainly for the full size and subcompact categories. Other discriminant variables , though weaker than the previous ones, indicate that carpoolers have worked longer at their present places of employement, are married rather than single, and have lived longer at their last residence. They are somewhat older, and have larger families. The following variables do not discriminate between the two groups: (1) number of persons in household with driver's license, (2) number of autos owned, (3) age of the car that is used for the work trip, (A) sex, (5) income, (6) professional status, and (7) education. Thus, the emerging picture of the typical carpooler in the Chicago area, in comparison with those who drive alone in their private automobile, is that the carpooler has a larger family, a larger car, has lived a longer time at his or her last residence, and has been working longer at the same place of employment. In short, the carpooler may be somewhat later in his or her life cycle than the solo driver. Table 2 shows also that auto ownership is not related to ridesharing practice: wh-fle carpoolers own on th^ average 1.8 cars per household, solo drivers own 1.9 cars per household, a slight and statistically insignificant difference. The average number of licensed persons per household is 2.2 in both groups. Travel characteristics of solo drivers and carpoolers are suimnarized in Table 3. Cost and time are subjective measures as reported by respondents in the survey. A multivariate test using Wilks Lambda criteria performed on seven variables showed that solo drivers differ significantly from carpoolers (F= 22.2, p <_ 0.001). The multivariate explained variance is larger than that of the socioeconomic characteristics (13.5% vs. 9.8%), The variables are ordered according to their discriminant CAR SIZE DISTRIBUTION ■— — CARPOOLERS (N=259) ""• • SOLO DRIVERS (N=328) ^ 10. # rSS .^ .# A- ^ ~^ CAR SIZE .^^^' <.^^ Figure 2 10 TRAVEL CHARACTERISTICS COMPARISON AMONG SOLO DRIVERS AND CARPOOLERS SOLO DRIVERS CARPOOLERS F-RATIO** PR0B.< EXPLAINED VARIANCE (%) MEAN^ N^ SD^ riEANj ^2 SD2 MULTIVARIATE STATISTICS MANOVA --- 382 ... 323 - 22.2 0.001 13.5 UNIVARIATE STATISTICS TOTAL COST, including gasoline and deprec- iation, driving alone one-way ($) GASOLINE COST, one-way ($) TRAVEL TIME ONE-WAY (minutes) TRAVEL TIME DRIVING ALONE (minutes) DISTANCE HOME-WORK (miles) DISTANCE TO NEAREST PUBLIC TRANSP. STA. (miles) WALK FROM CAR TO WORK (minutes) 1.27 0.54 26.5 26.5 11.2 3.7 2.3 337 347 380 380 376 263 342 0.94 0.42 14.8 14.8 9.1 6.6 2.3 1.75 0.84 34.3 32.2 16.3 3.7 3.1 271 285 321 318 319 215 294 1.71 0.60 16.8 15.9 12.9 7,5 2.0 58.0 54.4 40.8 31.1 37.1 0.0 3.1 0.001 0.001 0.001 0.001 0.001 !M.S. i^.S. 8.7 7.9 5.5 4.3 5.0 *N-HUMBER OF OBSERVATIONS SD-STANDARO DEVIATION N.S.-NOT SIGNIFICANT **-DEGREES OF FREEDOM: 7 AND 697 FOR THE MULTIVARIATE TEST 1 AND N., + N2-2 FOR EACH UNIVARIATE TEST T-ble ? power where th first variable, the tc al reported cost including gasoline and depreciation, explains 8.7% of the variance. As expected, if carpoolers would drive alone they would be expected to have a higher travel cost than solo drivers since carpoolers live farther from work. Note that carpoolers on the average spend 2.1 minutes more in ridesharing than if they would drive alone. The last two variables listed in Table 3 show that walk time from car to work and also the distance from home to the nearest public transportation station do not significantly differentiate the two groups. Because the cost, time, and distance measures are highly correlated, the multivariate explained variance does not equal the sum of the individual measures . 11 Figure 3 provides more detail on how carpoolers differ from solo drivers with respect to the distance to work. While the percentage of commuters in the range of 11 to 20 miles is rather similar for the two groups, this percentage differs somewhat in the less than 10 mile range and in the more than 20 miles work trips. DISTANCE TO WORK DISTRIBUTION ^■■■■iii""—™- CARPOOLERS (N=319) iiiiiiiiiiiiniiiiiiiSOLO DRIVERS (N=376) UJ o 1x1 '''*(pii"«*'***«»l»^ ~T r~i — ! 0-5 5-10 n-15 16-20 21-25 26-30 31 + DISTANCE HOME-WORK (MILES) Figure 3 12 A few comments are in order. First, a discriminant analysis (Appendix A) performed on both the demographic and travel characteristics showed that only 61.7% of the 705 commuters were correctly classified by the discriminant function. Since fay pure chance the expected correctly classified proportion is 50,., it follows that the demographic and travel characteristics add in only 11.7% of the cases, a small and negligible proportion. In summary the percentage of explained variance presented in Tables 2 and 3 and also not independently the results of the discriminant analysis indicate that demographic and travel characteristics are poor indicators of whether a commuter to work is driving alone or sharing a ride. Second, a comparison of the "explained variance" column of Tables 2 and 3 shows that the solo drivers and carpooling groups are better distinguished from each other by the travel characteristics than by socioeconomic characteristics. Note also that the socioeconomic variable that best distinguishes between the two groups is car size. This result is consistent with the declining role of socioeconomic variables in the explanation and prediction of consumer choice among relatively affluent middle class population (Yankelovich, 1964; Katona, 1975). Finally, the results are partially inconsistent with Alan M. Voorhees and Associates (1973) study of commuters on the Hollywood Freeway in the Los Angeles area. The only statistically significant discriminant variables in common with the present study and the Alan M. Voorhees study are distance to work and travel time. The earlier study, in contrast to the present one, found that carpoolers tend to be somewhat younger than solo drivers. This discrepancy between the two studies may be attributed to the small number of carpoolers , 108 > in the Alan M. Voorhees study and also to the different locations of the two studies. It is shown later, however, that attitudinal differences between carpoolers and solo drivers are found to be similar in the two studies and are perhaps more universal than the demographics and travel characteristics . 13 Ridesharlng Cognitive Profile Of the ten attributes of the cognitive profiles (Figure 1) only the attribute "Safe From Crime" was found not to differentiate the two groups or to correlate with any of the attributes. Table 4 presents the means and standard deviations of the ridesharing cognitive profile of the nine remaining attributes for solo drivers and carpoolers. Each attribute has been rated on a semantic scale from "1" to "7" where "1" means very low, "7" very high, and "4" serves as the neutral ground. A multivariate test performed on the whole vector of nine attributes showed that the two groups of respondents differ significantly (F- 30.6, p <_ 0.001). The univariate tests and the means displayed for convenience in Figure 4 lead to the following observations . First , solo drivers differ highly from carpoolers in the evaluation of ridesharing with respect to convenience, reliability, pleasure, comfort, and time (in this order) but do not differ in their evaluation of ridesharing with respect to cost, energy, traffic problems, and air quality. Note that the F-ratios and the explained variances are large when compared to the respective measures found for travel and socioeconomic characteristics. Second, solo drivers tend to evaluate carpooling on all nine attributes on the average at or near the middle ground "4" on the low side of the scale between "3" and "4". This result implies that solo drivers hold a neutral position toward ridesharing with a slight tendency to perceive it inconvenient, not reliable, etc. If solo drivers would have a dec rly negative attribute ^ rofile toward ridesharing one could not easily change their position; but it is suggested that given a general neutral position, a change in attitude might be achieved by advertisement and promotional means (for a discussion of the relation between neutral attitudes and attitude change, see Howard and Sheth, 1969, chapter 18) . Third, on the average carpoolers evaluate ridesharing as being clearly convenient, reliable, pleasant, comfortable, and economical. To a lesser extent they perceive ridesharing as time saving and low in creating traffic problems and pollution. In this context, it is noted that ridesharing cognitions of carpoolers and solo drivers measured by Alan M. Voorhees and Associates (1973, Figure 12), were compatible 14 EVALUATION OF RIDESHARING PROFILE VERY LOW 1 1 I r 7 VERY HIGH CARPOOLERS (N=323) •iumi MM SOLO DRIVERS (N=382) Figure A \ \ I ' ft / y r ^ X / M CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION II DIFFERENCES IN RIDESHARING EVALUATIONS AMONG CARPOOLERS AND SOLO DRIVERS (1=VERY LOW ; 7=VERY HIGH) MEANS F-RATIO** PR0B.< EXPLAINED VARIANCE SOLO DRIVERS (N= 382) CARPOOLERS (N- 323) MULTIVARIATE STATISTICS MAN OVA 30.6 0.001 28.4 UNIVARIATE S.ATI5TICS CONVENIENT 3.3 (1.7)* 5.1 (1.6) 197.6 0.001 22.0 RELIABLE 3.7 (1.5) 5.3 (1.5) 195.7 0.001 21.8 PLEASANT 3.9 (1.5) 5.3 (1.4) 162.7 0.001 18.8 COMFORTABLE 3.6 (1.5) 5.1 (1.5) 144.5 0.001 17.1 SAVES TIME 3.5 (1.6) 4.6 (1.7) 80.5 0.001 10.3 EXPENSIVE 3.1 (1.4) 3.1 (1.4) 0.3 N.S. ENERGY CONSUMING 3.7 (1.7) 3.9 (1.9) 1.9 N.S. --.-• TRAFFIC PROBLEMS 3.7 (1.5) 3.5 (1.7) 2.9 N.S. - . -• POLLUTION 3.8 (1.5) 3.7 (1.6) 0.6 N.S. ■STANDARD DEVIATIONS ARE GIVEN IN PARENTHESIS -DEGREES OF FREEDOM 9 AND 695 FOR MULTIVARIATE TEST 1 ".:iD 703 FOR EACH UNIVARIATE TEST Table 4 15 with those obtained herein in spite of the differences among the scales used in the two studies. The largest differences between carpoolers and solo drivers were found by Alan M. Voorhees in the following two bipolar semantic scales related to reliability; DON'T MIND RELYING r-, r-, r-, .-, r-, r-, .. DISLIKE RELYING ON OTHERS '■-' '-^ '-■' '--' "-J U LJ qn OTHERS DON'T MIND HAVING PEOPLE DEPEND ON ME [] [] [] [] [] [] [] DISLIKE HAVING PEOPLE DEPEND ON ME Figure 5 details differences between carpoolers and solo drivers in the evaluation of ridesharing with respect to convenience. The two distributions are distinctly different, particularly at the extreme points of the scale that is at 1, 2, 6 and 7. Note that 22% of the variance is explained by the classification of respondents into the two groups (Table 4) . DISTRIBUTION OF RIDESHARING CONVENIENCE CARPOOLERS (N= 323) iko.i.MMML.iiSOLO DRIVERS (N=382) CD o (VERY LOW) 12 3 4 5 6 7 (VERY HIGH) EVALUATION OF RIDESHARING CONVENIENCE Figure 5 16 An additional measure of attitudinal differences between the two groups of respondents based on the carpooling attributes has been obtained through a discriminant analysis presented in Appendix B. The discriminant function was able to correctly classify 73.6% of the respondents, that is, 23.6'' in addition to the 50% that are expected to be classified correctly by random assignments to groups, or about twice the discrimination beyond random that could be achieved by socioeconomic and travel character- istics. Structuring Cognition Insight to the latent psychological dimensions by which respondents evaluate ridesharlng can be achieved by inspecting the correlations between the nine evaluated attributes. There is no 'reason to expect independence between the nine attributes. The attributes were chosen for inclusion in this survey on the basis of their potential importance in discovering the latent psychological dimensions. Table 5 presents the correlations between the ridesharlng attributes as evaluated by carpoolers. Note that two clear subsets of attributes are formed. It is interesting that those first five attributes that best discriminate between the two groups (convenience, reliability, etc.) are interrelated but not related to the other attributes which are however themselves interrelated. Principal component analysis with varimax rotation (Harman, 1967) was applied to the correlation matrix. The number of factors retained was determinei by a comparison of the set of eigenvalues obtained from analysis of random data matrices of the same order as the actual data matrix and by consideration of the "Kaiser rule" in which eigenvalues greater than one are retained (Horn, 1965) . Either of these criteria resulted in the selection of two factors. The largest three eigenvalues, in decreasing order, were 3.3, 2.2, and 0.8. Table 6 shows the two factors comprising the nine ridesharlng attributes. The "Factor" column gives the subjective label for each factor. The factor loadings are those significantly different from zero. The "Explained Variance" column lists the percent variance of each 17 CARPOOLERS : RAW CORRELATIONS AMONG RIDESHARING EVALUATIONS Table 5 CARPOOLERS: FACTOR STRUCTURE OF RIDESHARING EVALUATIONS FACTORS (% VARIANCE EXPLAINED ) ATTRIBUTE FACTOR LOADIIG EXPLAINED VARIANCE (%) IN FACTOR IN THE OTHER FACTOR I TIME- CONVENIENCE (37.2) CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME 0.85 0.83 0.84 0.80 0.76 72.2 68.9 70.3 63.8 57.5 0.1 0.1 0.2 0.0 2.7 II PRIVATE AND PUBLIC COST (24.8) EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION 0.58 0.70 0.83 0.84 33.2 49.4 68.9 70.3 0.0 0.1 0.1 0.1 Table 6 18 attribute which is accounted for by the factor in question (the square of the respective loading) , and the percent which is accounted by the other factor. These two pieces of information (which sum to the total percent of the attributes variance accounted for by both factors, or che communality) indicate the strength and uniqueness of the attribute factor relationship, respectively. The "Factor" column includes also the percentage of the variance in the corresponding attribute set which is accounted for by this factor and equals the average in factor variance (over all nine attributes). The very low figures in the "Other factor" column and the relative high figures in the "In factor" column, shov; that each of the two factors is strong and unique. Two revealing observations result from the factor analysis. First, the grouping of time with such qualitative attributes of convenience, comfort, etc., was unexpected. Indeed, the traditional approach in transportation research is to separate between time and cost on one hand and qualitative aspects on the other. However, the characteristics of ridesharing and, by comparison, also of solo driving, are related to a variety of time aspects, such as : fixed or flexible schedules, spending time to pick, up other riders, spending time to wait for other riders, relying on others to be on time, additional time required for errands, etc. The second insight to the latent psychological dimensions is the inclusion of personal cost ("expensive") in the same factor with the public cost attributes of energy, traffic, and pollution. The term "Private" in the label of Factor II is preferred over "Individual" because the cost typically involves the household rather than only the individual. A factor analysis of ridesharing evaluations by solo drivers is presented in Table 7, The factor structure is similar to that obtained for carpoolers, but a comparison with Table 6 shows, however, that the percentage of variance explained by each factor is lower than for the carpoolers sample. Note also that the factor loadings and the "In Factor" explained variance for eight of the nine attributes are lower than for 19 SOLO DRIVERS: FACTOR STRUCTURE OF RIDESHARING EVALUATIONS FACTORS {% VARIANCE EXPLAINED ) ATTRIBUTE FACTOR LOADIh'Q EXPLAINED VARIANCE (%) IN FACTOR IN THE OTHER FACTOR I TIME- CONVENIENCE (35.3) CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME 0.79 0.79 0.80 0.79 0.77 63.2 61.9 63.4 62.1 58.8 2.0 0.0 0.4 0.0 2.8 II PRIVATE AND PUBLIC COST (21.0) EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION -0.45 -0.69 -0.80 -0.77 20.6 47.0 64.5 59.9 0.2 0.3 0.0 0.3 Table 7 the carpoolers sample. These results suggest first that familiarity with an object of attitude (in this case, carpoolers with ridesharing) is enhancing their significance; second, that lack of familiarity with an object (solo drivers with ridesharing) is increasing noise (error) in the data. An additional, but not independent, interpretation of the factor structure, is that Factor I captures the perceived disadvantages of ridesharing, while Factor II is associated with the perceived advantages in comparison to the drive-alone mode. This interpretation is substantiated by the drive-alone evaluations presented below. Drive-Alone Cognitive Profile These same nine attributes were also rated in the context of the drive alone mode as Illustrated in Figure 1. The row means, standard deviations, and the statistical tests performed on the drive-alone means are displayed in Figure 6 and the evaluations are presented in Table 8. 20 EVALUATIONS OF DRIVE- ALONE PROFILE I CARPOOLERS (N=323) n.im iiimitSOLO DRIVERS (N=382) VERY LOW 1 7 VERY HIGH o —I < — iti- CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION II Figure 6 DIFFERENCES IN DRIVE-ALONE EVALUATIONS AMONG CARPOOLERS AND SOLO DRIVERS (1=VERY LOW; 7=VERY HIGH) MEANS F-RATIO** PROB < EXPLAINED VARIANCE {%) SOLO DRIVERS (N= 382) CARPOOLERS (N= 323) MULTIVARIATE STATISTICS MANOVA 10.4 0.001 11.8 UNIVARIATE STATISTICS CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME 6.6 (1.2)* 6.5 (1.2) 5.9 (1.5) 6.3 (1.4) 6.4 (1.3) 6.2 (1.5) 6.2 (1.4) 5.3 (1.8) 5.7 (1.6) 5.9 (1.6) 15.0 8.7 23.9 26.0 26.7 0.001 0.003 0.001 0.001 0.001 2.1 1.2 3.3 3.6 3.7 EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION 4.7 (1.8) 4.8 (2.2) 5.1 (1.9) 5.1 (1.9) 5.6 (1.7) 5.0 (2.2) 5.2 (2.0) 5.3 (1.^) 46.7 1.8 0.8 1.1 0.001 N.S. N.S. N.S. 6.2 *-STANDARD DEVIATIONS ARE GIVEN IN PARENTHESIS **-DE6RE£S OF FREEDOM: 9 AND 695 FOR THE MULTIVARIATE TEST 1 AND 703 FOR EACH UNIVARIATE TEST Table 8 21 A multivariate test performed on the vector of nine attributes showed that the two groups differ significantly but to a lesser degree than in the case of the ridesharing evaluation (F= 10.4, p £ 0.001). An inspection of the individual means and the univariate tests leads to two main observations. First, both groups of commuters perceive the dri«re-alone to work mode hi^h on the qualitative attributes of convenience, reliability, comfort, and also on saving time. Second, solo drivers are somewhat more positive toward their o^jn mode of trans- portation than carpoolers are toward driving alone. This difference is statistically significant for all attributes with the exception of the public cost attributes of energy, traffic, and pollution. It should be noted that the explained variance is small in spite of the statistical significant results. Of special interest is the interrelation of the attributes, that is, the latent psychological dimensions by which respondents evaluate the drive-alone mode. It is remarkable that in spite of the pronounced differences between the evaluations of the two modes of travel as seen by a comparison of Figures 3 to 5, the drive alone evaluation factors (Tables 9 and 10) are virtually identical to the previously described ridesharing factors. A comparison between Tables 9 and 10 suggests that the time -convenience factor explains more variance for solo drivers than for carpoolers (41.5% vs. 38.0% respectively). On the contrary, the cost factor explains more variance for carpoolers than for solo drivers (26.2% vs. 24.8% respectively). These results illustrate the high weight of the psychological time-convenience factor in the solo drivers cognition of driving alone. Differences Among Ridesharing and Drive-Alone Cognitive Profiles The attribute evaluations presented above were measured with respect to ridesharing and separately for the drive-alone mode. To get a more comprehensive grasp of the ridesharing cognition and to relate it to both the affective and the intentional components, when the drive— alone mode serves as a baseline, consideration is given to the difference 22 CARPOOLERS: FACTOR STRUCTURE OF DRIVE-ALONE EVALUATIONS FACTORS {% Variance explained ) ATTRIBUTE F XTOR LOADING EXPLAINED VARIANCE (%) IN FACTOR IN THE OTHER FACTOR I TIME- CONVENIENCE (38.0) CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME 0.85 0.78 0.76 0.82 0.85 72.8 60.1 57.7 67.7 71.8 1.6 0.6 0.2 0.0 0.3 II PRIVATE AND PUBLIC COST (26.2) EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION -0.58 -0.72 -0.89 -0.87 33.5 51.8. 79.2 76.5 3.2 0.2 0.0 0.0 Table 9 SOLO DRIVERS: FACTOR STRUCTURE OF DRIVE-ALONE EVALUATIONS FACTORS {% VAiUANCE EXPLAINED ) ATTRIBUTE FACTOR LOADING EXPLAINED VARIANCE (%) IN FACTOR IN THE OTHER FACTOR I TIME- CONVENIENCE (41.5) CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME 0.90 0.90 0.76 0.87 0.86 81.2 80.6 57.3 75.9 74.7 0.0 0.2 2.2 0.6 0.1 II PRIVATE AND PUBLIC COSTS (24.8) EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION -0.54 -0.70 -0.85 -0.84 29.1 49.4 71.5 69.7 1.8 1.2 0.8 0.2 Table 10 23 between the drive- alone and rldesharing evaluations as a measure of evaluation on each attribute. This difference is computed by subtracting the individual measures summarized in Table 4 from those In Table 8, and will be denoted by 6 . , 1=1,..., 9, where i~ ^i, drive- alone "'^i, ridesharing and X. is the evaluation of the attribute i on the corresponding i, mode mode. The 6. measures and the respective standard deviations are presented in Tables 11 and 12 for solo drivers and carpoolers, respectively. The hypothesis that the 6 . measures are not different from zero has been rejected for both solo drivers and carpoolers for all attributes with the exception of "pleasant" in the carpoolers group. F-ratios DIFFERENCES AMONG DRIVE-ALONE AND RIDESHARING EVALUATIONS FOR SOLO DRIVERS (&. MEASURES) FACTOR MEAN DIFFERENCE F-RATIO** PROB. < EXPLAINED VARIANCE (%) MULTIVARIATE STATISTICS MANOVA 144.2 0.001 77.5 UNIVARIATE STATISTICS CONVENIENT 3.3 (2.2)* 873.5 0.001 57.6 RELIABLE 2.8 (2.0) 763.9 0.001 52.1 I PLEASANT 2.0 (2.2) 297.3 0.001 29.7 COMFORTABLE 2.7 (2.2) 570.8 0.001 44.8 SAVES TIME 2.9 (2.2) 717.2 0.001 50.5 EXPENSIVE 1.6 (2.3) 167.6 0.001 19.3 ENERGY CONSUMING 1.1 (2.9) 53.4 0.001 7.1 II TRAFFIC PROBLEMS 1.4 (2.3) 136.3 0.001 16.2 POLLUTION 1.3 (2.1) 166.4 0.001 19.1 *-STANDARD DEVIATIONS ARE GIVEN IN PARENTHESIS **-DEGREES OF FREEDOM : 9 AND 373 FOR THE MULTIVARIATE TEST 1 AND 381 FOR EACH UNIVARIATE TEST (N=382) Table 11 24 DIFFERENCES AMONG DRIVE-ALONE AND RIDESHARING EVALUATIONS FOR CARPOOLERS ( h. MEASURES) FACTOR MEAN DIFFERENCE F-RATIO** PR0B.< EXPLAINED VARIANCE (%) I MULTIVARIATE STATISTICS MANOVA 64.8 0.001 65.0 UNIVARIATE STATISTICS CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME 1.1 (2.2)* 0.9 (2.0) 0.0 (2.2) 0.5 (2.1) 1.3 (2.1) 81.5 67.2 0.2 26.3 117.8 0.001 p. 001 N.S. 0.001 0.001 10.4 8.7 3.6 14.4 II EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION 2.5 (2.3) 1.1 (3.2) 1.7 (2.5) 1.6 (2.3) 397.6 40.1 157.6 160.4 0.001 0.001 0.001 0.001 36.1 5.4 18.3 18,6 *-STANDARD DEVIATIONS ARE GIVEN IN PARENTHESIS **-DEGREES OF FREEDOM : 9 AND 314 FOR THE MULTIVARIATE TEST 1 AND 322 FOR EACH UNIVARIATE TEST (N=323) Table 12 and the percentage of explained variance for solo drivers show that the perceived differences between ridesharing and driving alone are very pronounced (compare also Figures 4 and 6) especially for the time-convenience factor, reinforcing those results obtained from the drive-alone cognitive profile. Factor analysis applied to the 6 measures yielded factors similar to those obtained by the individual measures and are presented in Appendixes C and D. 25 Affect Toward Ridesharing and the Intention to Share a Ride Figure 7 presents the affect distribution toward ridesharing for the solo drivers and the carpoolers groups. The affect was measured by the answer to the question "All things considered, which statement best describes how you like the idea of yo , being a member of a carpool?" (See Figure 1) . There is little need for a statistical test to detetmine the two groups are highly differentiated by the affect measure. Solo drivers are split along the continuum from "Like Extremely" to "Dislike Extremely", with about 20% of the solo drivers being neutral, while almost all carpoolers are positive toward ridesharing. Figure 8 displays the carpooling intention distribution for solo drivers. The intention was measured by the answer to the question "How likely are you to join a carpool within the next two or three months?" Less than 10% of the 376 solo drivers who answered the question stated a positive intention. The intention measure is most appropriate for the prediction of behavior. The results suggest that under present conditions only a small percentage of solo drivers intend to carpool regularly in the immediate future. However, the overall prediction of future trends in ridesharing is quite complex because some of the present carpoolers are likely to switch back to the drive-alone mode. This statement is based on the fact that about A0% of the present solo drivers surveyed in this study reported that they had carpooled in the past on a regular basis for an average period of two years, out discontinued carpooling. The relationship between affect and intention on one side and the latent cognitive factors of time-convenience and private-and-public- cost on the other side, will be explored in the next section. Models Relating Cognition Factors to Affect and Intention It was shown that there are two latent factors underlying the cognition of ridesharing, drive-alone, or the difference between them. The time-convenience factor was interpreted as the perceived negative evaluation of ridesharing while the cost (private and public) factor as the perceived positive evaluation. Researchers in both social psychology and consumer psychology have theorized that there is a linear additive relationship between evaluations (cognition) and between affect and intention (Fishbein, 26 AFFECT TOWARD RIDESHARING iCARPOOLERS (N=321) iiiiMiii SOLO DRIVERS (N=382) LIKE EXTREMELY iiiiiiitiii LIKE MODERATELY iiiiiiiiriiiiiiii iiiiiiiiKigiiii LIKE SLIGHTLY NEITHER LIKE NOR DISLIKE m»ii" DISLIKE SLIGHTLY DISLIKE MODERATELY DISLIKE EXTREMELY iiiiiiiii iiKiiaiii iiityiii '■■iiiiiriaitii iiiiiiii iiiiiiiiiiiiiiii 10 15 20 25 PERCENT 30 35 40 45 Figure 7 INTEfJTlOn TO CARPOOL DISTRIBUTIOf! FOR PRESENT SOLO DRIVERS DEFINITELY '.•JILL > VERY LIKELY - SOMEI^HAT LIKELY — CANriOT SAY — SOMEWHAT UNLIKELY VERY UNLIKELY DEFINITELY WILL NOT 1 ■ 13 20 30 40 50 50 PERCE.'IT N = 376 Figure 8 27 1967; Sheth, 1974). A linear additive relation implies that positive and negative evaluations compensate for one another. Recently, however, several researchers have expressed the concern that a linear additive presumption may be a serious limitation to understanding attitudinal structure (Day, 197:; Raju and Sheth, 1974), Horowitz suggests an attitudinal ridesharing model that allows for noncompensatory relation: "Is it possible that evaluations interact among themselves so that a negative evaluation can reduce the intention to carpool regardless of the magnitude of the positive evaluation?" (1975, p. 3). For the purpose of describing the following assume that each individual is rated as either "High" or "Low" on each of the two factors according to whether his or her respective factor scores are higher or lower than the average score. (One could divide the continuum into more than two parts but for model testing it is sufficient to have two categories). Then, each group (carpoolers and solo drivers) will be segmented into four subgroups according to the combination of the two factors, as shown in Figure 9. T denotes the time-convenience factor, and C the private-and-public-cost factor. Consideration of the meaning of the two factors in relation to ridesharing and solo driving results in the following interpretation of the cells. Cell {1,2} includes those individuals who are more positive than the average toward ridesharing along both factors, cell {2,1} includes those individuals who are negative toward ridesharing on both factors, while the other two cells include the obvious combinations of positive and negative factor scores. Following the notation introduced above, and taking the position that affect is determined by the factors T and C, a linear-interactive model for affect is: 28 A SEGMENTATION BASED ON THE COGNITIVE PROFILE LOW HIGH (NEGATIVE*) (POSITIVE) LOW (POSITIVE) HIGH (NEGATIVE) [1,1] [1,2] [2,1] [2,2] where ♦"NEGATIVE" AND "POSITIVE" TOWARD RIDESHARING Figure 9 ^j.i = W -T,+C.+Y. ,+e. ., ijk "^ i 3 'xj ijk A. .,= individual k's affect toward ridesharing, where his or her T-factor score is i (low, high) and his or her C-factor score is j (low, high) mean affect over all four cells the contribution of factor T to affect at level i the contribution of factor C to affect at level j interaction between the T. and C, levels 1 j individual k's error in cell {i,j} 29 An ordinary 2x2 analysis of variance (ANOVA) can be used to test the model. The use of ANOVA depends on the statistical assumption that e ,, are independent random variables normally distributed with constant variance. In the present application of ANOVA these statistical assumptions pose no problem because the number of observations is relatively large. Use of ANOVA requires independence between observations. Hence, it is necessary that different individuals belong in different cells. This assumption is clearly satisfied in the present design. The ANOVA allows simple, powerful tests for each of the T., C . , and y.. terms separately. A similar model can be written for intention, that is. where I. ^, denotes individual's k intention to share a ride and all other terms are similar to those in the affect model but refer to intention. Test of the Affect Model Based on the segmentation discussed above each respondent has been assigned to one of the four cells according to his or her factor scores, T and C. Note that "Low" and "High" are relative to the weighted (by factor loadings) average difference between the ridesharing and drive alone evaluations. Figure 10 pre- ents the affect means f r each cell for sclo drivers and carpoolers separately. The corresponding standard deviations and cell sizes are included in ^pendix E. Two main results emerge from Figure 10. First, the time-convenience factor, that is, whether a respondent is categorized into "Low" or "High" on T, is related to his or her affect to a larger extent than is the factor C. This is seen from a comparison of the slopes of the lines to the distance between the lines for the carpoolers and solo drivers groups, separately. Second, that the 30 lines are non-parallel suggests an interaction between the factors, especially for solo drivers. THE RELATION BETWEEN THE AFFECT TOWARD RIDESHARING AND THE COGNITIVE FACTORS . (LIKE EXTREMELY) 7 6 _ CD a: .) ■ ,- ' ?"■•■- .-I.:.. :'.\:.,l variables (Table 2) and the "distance" variable have been tested. Tables 14 and 15 enumerate only those variables for which at least one main contribution, T or C, was found to be significant at the level p <_ 0.05. The results show that there are significant socioeconomic differences among the four cells. First, solo drivers who are more positive toward ridesharing than the average with respect to factor T (cells {1,1} and {1,2}) are from larger households, have worked a shorter time at their last place of employement and have lived at their present residence a shorter time than the other solo drivers. Second, those solo drivers who are more positive toward ridesharing on factor C (cells {1,2} and {2,2}) typically live farther from their work, are males from households with more driving licences, have a higher education, income, and occupation level than the other solo drivers. The emerging picture of the ridesharing target market, that is, cell {1,2}, is that it includes employed individuals with high socio- economic status, as measured by education, income, and occupation, are from relatively large households and have worked and lived at their last place of employement and residence, respectively, for a shorter time than the other solo drivers. It is suggested, then, that these types of individuals art sensitive to the privata-and-public-cost of solo driving in spite of the fact that they are now driving to work alone. A ridesharing promotional campaign could address this segment of people with issues related to both factors T and C. It is also suggested that the optimal strategy toward all other types of cotomuters, the large majority of solo drivers, is to concentrate on issues related to the time-convenience factor and to ignore the cost related advantages of ridesharing. 35 THE DISTRIBUTION OF THE SOLO-DRIVERS SOCIOECONOI^IC VARIABLES IN THE COGNITIVE FACTORIAL DESIGN CELL [1,1] [1.2] [2,r [2.2] VARIABLE MEAN N SO MEAN N SD MEAN N SO MEAN N SD YEARS AT PRESENT EMPLOYER 7.6 94 7.8 6.7 87 8.2 8.3 104 9.5 10.5 89 10.3 YEARS AT PRESENT RESIDENC-E 6.7 95 5.9 5.8 87 6.1 7.2 105 7 3 8.8 88 8.1 HOUSEHOLD SIZE 3.3 92 1.6 3.3 84 1.4 2.7 104 1.3 3.0 86 1.5 NUMBER OF LICENSED DRIVERS IN HOUSEHOLD 2.2 95 1.1 2.3 87 1.1 2.0 106 0.9 2.4 90 1.1 HOUSEHOLD ANNUAL INCOME* 8.7 83 3,0 9.0 76 3.3 8.3 89 3.2 9.8 83 2.4 OCCUPATION* 2.5 96 1.2 2.2 85 1.2 ?-A. 106 1.1 2.4 90 1.1 EDUCATION* 4.9 95 1.1 5.4 85 1.1 4.7 106 1.3 5.2 91 1,0 SEX (I=MALE. 2=FEMALE) 1.5 95 0.5 1.3 87 0.5 1.6 108 0.5 1.4 91 0.5 DISTANCE HOME-WORK (MILES) 10.3 95 7.7 13.4 84 9.9 10.0 107 9.3 11.7 90 9.1 * FOR UNITS, SEE TABLE 1 . Table lA ANOVA ANALYSES* ON SOCIOECONOMIC AND DISTANCE VARIABLES- SOLC DRIVERS VARIABLES FACTOR T FACTOR C DEGREES OF FREEDOM F-RATIO PROS. F-RATIO PROB. YEARS AT PRESENT EMPLOYER 5.2 0.023 — li370 YEARS AT PRESENT RESIDENCE 5.1 0.024 — 1;371 HOUSEHOLD SIZE 7.4 0.007 — l;362 NUMBER OF LICENSED DRIVERS IN HOUSEHOLD 5.7 0.018 1;374 HOUSEHOLD ANNUAL INCOME — 8.2 0.005 U327 OCCUPATION — 5.5 0.020 U373 EDUCATION — 20.6 0.001 U373 SEX -_- 11.6 0.001 1;377 DISTANCE, HOME-WORK — 5.4 0.012 l;372 * NO INTERACTION TxC WAS FOUND SIGNIFICANT Table 15 36 DISCUSSION This study demonstrates how attltudinal measures and the resulting factorial dimensions and segmentations can provide results that bear on a range of possible promotional strategies regarding ridesharing. Carpooling promotion could make use of the normative social influence of regular carpoolers. Research in consumer research (for example, Bunkrant and Cousineau, 1975) shows that people use others' product evaluation as a source of information about the product. In particular, after observing others evaluating a product favorably, individuals, in general, perceive the product more favorably than they would have in the absence of this observation. Since the attitudes of carpoolers toward ridesharing are very positive, this information could be communicated to solo drivers for the promotion of ridesharing. The most significant result of the present study is that the intention to carpool and the affect toward ridesharing are related to the perceived negative evaluations rather than to the positive aspects. It is suggested, then, that ridesharing campaigns should address the perceived negative evaluations related to the time-convenience factor. A possible approach to resolving negative perceptions may be guided by the psychological theory of cognitive dissonance (Festinger, 1957) . The hypothesis is that a significant difference between an individual's own point of view and some available information, say, promotion of ridesharing by emphasizing its advantages, gives rise to an uncomfortable state of dissonance which can be reduced either by changing one's own position or by rationalizing the available information. New information, based on proven facts and presented in an effective manner, according to the theory of cognitive dissonance, may result in a change of attitude. Thus information for the promotion of ridesharing should differ substanti- ally from the prevalent attitudes, giving rise to dissonance. Ways of transmitting the message that ridesharing is convenient, reliable, comfortable, and saves time should be specifically studied. The results suggest also that appeals to private economy and public interest issues of energy conservation and traffic and pollution reduction have only a slight chance of changing attitudes toward carpooling; such 37 appeals are not likely to create dissonance and hence may not induce solo drivers to change their travel behavior. Other results, based on a segmentation technique, showed, however, that males with a relative high socioeconomic status are more sensitive to private economy and public interest issues the a are other individuals. Future Research The carpooling data set employed in the present study may also help to answer the following two questions: First, what is the sensitivity of the choice between ridesharing and drive-alone to policies such as preferential lanes for carpoolers or changes in gasoline price? A second question, of interest in consumer psychology methodology in general and also for the forecast and promotion of ridesharing in particular, would be a comparison of attitudinal and socioeconomic characteristics of solo drivers who were regular carpoolers in the past but discontinued with those who never carpooled. 38 ACKNOI'ILEDGEMENT The authors wish to thank Dr. Wilfred W. Recker of Research Laboratories, General Motors Corporation for a critical reading of the paper. 39 REFERENCES Anderson, N. H. Scales and Statistics: Parametric and Nonparametric. Psychological Bulletin , 1961, 58, 305-316. Barkow, B. Carpooling, The Worm's Eye Vi ew. Paper presented at the American Psychological Association annual Convention, Washington, D. C, 1976. Ben-Akiva, M. E. and Atherton, T. J. Choice Model Predictions of Carpool Demand: Methods and Results . Paper presented at the American Psychological Association Annual Convention, Washington, D. C, 1976. Berry, W. L. On the Economic Incentives for Commuter Carpooling . Ph.D. Thesis, Graduate School of Business Administration, Harvard University, Cambridge, Massachusetts, 1975. Bettman, J. R. , Capon, N., and Lutz, R. J. Cognitive Algebra in Multi- Attribute Attitude Models, Journal of Marketing Research , 1975, 12, 151-164, Blankenship, D. P. Utilizing Focus Group Dynamics to Ascertain Rules for Social Interaction for Carpoolers , Orange County Transit District, Santa Ana, California, 1975. Bunkrant, R. E. and Cousineau, A. Informational and Normative Social Influence in Buyer Behavior. Journal of Consumer Research , 2, 1975, 206-215. Carnegie-Mellon University, School of Urban and Public Affairs. An Evaluation of the SPRPC Carpool Public Transit Program , Report prepared for the Southwestern Pennsylvania Regional Planning Commission, Pittsburgh, Pennsylvania, 1975. Davis, F. W. et al. Ridesharing and the Knoxville Commuter , Report No. TCUT-1-75. Prepared for the Office of Environmental Affairs, U. S. Deparr-nent of Transportation. Day, G. S. Evaluating Models of Attitude Structures, Journal of Marketing Research , 1972, 9, 279-286. Dueker, K. J. and Levin, I. P. Carpooling; Attitudes and Participation . Paper presented at the American Psychological Association Annual Convention, Washington, D. C, 1976, Festinger, L. A Theory of Cognitive Dissonance . Stanford: Stanford University Press, 1957. FHWA, Department of Transportation, Federal Highway Administration News Letter 49-i May 22, 1975. Fishbein, M. Attitude and the Prediction of Behavior. In Fishbein, (ed.) Readings in Attitude Theory Measurement , New York: Wiley, 1967. 40 ■!:^itiSi Harman, H. H. Modem Factor Analysis . Chicago: University of Chicago Press, 1967. Hays, W. L. Statistics , New York: Holt, Rinehart, and Winston, 1963. Herman R. and Lam, T. Carpools at Large Suburban Technical Center. Transportation Engineering Journal , 1975, 101 , 311-319. Horn, J. L. A Rationale and Test for the Number of factors in Factor Analysis. Psychometrika . 1965, 30, 179-185. Horowitz, A. D. An Attltudinal Model of Carpooling Behavior . Research Laboratories, General Motors Corporation, Research Publication GMR-1969, August, 1975. Howard, J. A. and Sheth, J. N. The Theory of Buyer Behavior . New York: Wiley, 1969. Katona, G. Psychological Economics . New York: Elsevier, 1975, Kendall, D. C. Carpooling; Status and Potential , Final Report No. DOT-TSC-OST-75-23, Transportation Systems Center, U. S. Department of Transportation, Cambridge, Mass., June 1975. Krishnan, K. S. and Clelland, R. C. Selection of Undergraduate Freshmen using Discriminant Analysis. The Journal of Experimental Education , 1973, 41, 28-36. Margolin, J. B. and Misch, M. R. Incentives and Disincentives to Rldesharing . Paper presented at the American Psychological Annual Convention, Washington, D. C, 1976. Morrison, D. F. Multivariate Statistical Analysis . New York: McGraw-Hill, 1967. Raju, P. S., ard Sheth, J, N. Nonline ' r, Noncompensatory Relationships in Attituae Research , Working Paper Number 176, College of Commerce and Business Administration, University of Illinois, April 1974. Rosenberg, M. J. A Structural Theory of Attitude Dynamics, Public Opinion Quarterly , 1960, 24, 319-340. Rosenbloom, S., and Shelton, N. J. Carpool and Bus Matching Program for the University of Texas at Austin , Research Report 11, The Graduate Program in Community and Regional Planning, University of Texas at Austin, September, 1974. Sheth, J. N. A Field Study of Attitude Structure and the Attitude-Behavior Relationship. In J. N. Sheth (Ed.) Models of Buyer Behavior , New- York: Harper and Row, 1974. United States Congress, Emergency Highway Energy Conservation Act, Public Law 93-239, January, 1974. 41 (Alan M.) Voorhees and Associates, Inc. A Study of Techniques to Increase Commuter Vehicle Occupancy on the Hollywood Freeway . November, 1973. Zerega, A. M. and Ross, R, B. Application of Conjoint Measurement Technique s In Evaluating Carpooling Policies . Paper presented at the American Psychological Assoclr tlon Annual Convention, Washington, D. C. , 1976. Yankelovich, D. New Criteria for Market Segmentation. Harvard Business Review , 1964, 42, 83-90. 42 Appendix A DISCRIMINANT ANALYSIS BETWEEN SOLO DRIVERS AND CARPOOLERS - SOCIOECONOMIC AND TRAVEL CHARACTERISTICS i VARIABLE F-VALUE d.f. C^SOLO i aCP i 1 DISTANCE, HOME-WORK 37.1** 1;703 0.10 0.15 2 CAR SIZE 22.2** 1;702 2.90 3.26 3 SEX 9.0** 1;701 14.18 15.12 4 YEARS AT PRESENT EMPLOYER 12.3** 1;700 0.16 0.19 5 OCCUPATION 5.8* 1;599 4.59 4.42 6 HOUSEHOLD SIZE 5.8* 1 ; 698 1.21 1.37 7 MARITAL STATUS 5.2* 1;6 97 9.50 10.03 (CONSTANT) — -52.28 -62.71 F BETl^EEN GROUPS 14.2** 7; 697 ** : p<0.01 * : p<0.05 CLASSIFICATION INTO GROUPS ACTUAL CLASSIFIED AS SOLO CARPOOLER SOLO CARPOOLER 123\^1^0^ TOTAL 382 323 705 CORRECTLY CLASSIFIED : 61.7 % SOLO : 64.1 % CARPOOLERS : 58.8 % 43 Appendix B DISCRIMINANT ANALYSIS BETWEEN SOLO DRIVERS AND CARPOOLERS - ATTITUDE TOWARD RIDESHARING i VARIABLE F VALUE d.f. OfSOLO i < 1 CONVENIENT , 197.6** 1 -, 703 0.12 0.52 2 RELIABLE 38.9** 1;702 0.33 0.68 3 PLEASANT 10.8** 1;701 1.43 1.64 4 SAVES TIME 4.1* 1; 700 0.11 -0.04 (CONSTANT) --- -13.60 -17.50 F BETWEEN GROUPS 66.6** 4; 700 — — ** : p<0.001 * : o<0.05 CLASSIFICATION INTO GROUPS ACTUAL CLASSIFIED AS SOLO CARPOOLER SOLO CARPOOLER TOTAL 382 323 705 CORRECTLY CLASSIFIED : 73.6 % SOLO : 73.3 % CARPOOLERS : 74.0 % 44 Appendix C SOLO DRIVERS: FACTOR STRUCTURE OF THE 6: MEASURES FACTOR (% VARIANCE EXPLAINED ) AHRIBUTE FACTOR LOADING EXPLAINED VARIANCE (%) IN FACTOR IN THE OTHER FACTOR I TIME- CONVENIENCE (38.3) CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME 0.86 0.84 0.78 0.82 0.84 73.6 69.8 61,0 67.2 71.4 0.5 0.0 0.7 0.3 0.7 II PRIVATE AND PUBLIC COST (23.8) EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION 0.55 0.70 0,83 0.80 30.7 49.5 68.5 63.6 0.0 0.4 0.7 0.2 45 j;- Appendix D CARPOOLERS : FACTOR STRUCTURE OF THE h MEASURES FACTOR (% VARIANCE EXPLAINED ) ATTRIBUTE FACTOR LOADING EXPLAINED VARIANCE (%) IN FACTOR IN THE OTHER FACTOR I TIME- CONVENIENCE (36.8) CONVENIENT RELIABLE PLEASANT COMFORTABLE SAVES TIME 0.86 0.80 0.79 0.80 0.81 73.8 64.1 63.0 64,1 64.9 0.4 0.0 0.5 0.2 0.5 II PRIVATE AND PUBLIC COSTS (26.6) EXPENSIVE ENERGY CONSUMING TRAFFIC PROBLEMS POLLUTION 0.57 0.73 0.88 0.87 32.8 53.0 76.8 75.3 0.3 0.0 0.0 0.4 46 AFFECT INTENTION Appendix E AFFECT AND INTENTION DISTRIBUTIONS IN THE COGNITIVE FACTORIAL DESIGN * >-<^ (1) LOW (2) HIGH (1) LOW MEAN = SD = N = 4.2 1,8 95 MEAN = 4.9 SD = 1.7 N = 87 (2) HIGH MEAN = SD = H = 3.0 1.8 108 MEAN = 3.0 SD = 1.7 N = 91 '^^X. (1) LOW (2) HIGH (1) LOW MEAN = SD = N = 2.5 1.4 96 MEAN = 2.8 SD = 1.4 N = 87 (2) HIGH MEAN = SD = N = 1.8 0.9 108 MEAN = 2,2 SD = 1,2 N = 91 *N : NUMBER OF INDIVIDUALS SD : STANDARD DEVIATION 47 AM'