Tranportation (2013) 40:755 771 DOI 10.1007/11116-013-9465-6 Deciion maker and ocializer, ocial network and the role of individual a participant Kathleen Deutch Kontadino G. Goulia Publihed online: 1 May 2013 Ó Springer Science+Buine Media New York 2013 Abtract Model explaining and predicting human travel behavior have gone through many change in the pat few decade. A reearcher attempt to explain more and predict with more accuracy, the incluion of ocial interaction in modeling and imulation i being recognized a a neceity. Among thee effort, reearcher have focued on iue uch a the compoition of ocial network, and the contraint and influence that other have on patial deciion. An important apect that ha been undertudied however i the variability or heterogeneity of individual both a ocial network member and a participant in thee ocial network. Undertanding the role individual play in deciion-making in different ocial network can further define our model to include more accurate repreentation of human behavior. Thi reearch explore the difference between ocial network compoition, and the deciion role member play within different ocial network pecifically when deciding where to participate in activitie. A urvey wa conducted in Santa Barbara, California on ocial network involvement, network attribute and deciionmaking role within each network. Two eparate latent cla cluter analyi model were developed to claify ocial network involvement and role. Reult how that there are clearly different type of ocial involvement and role within network. Further data collection and analyi will be ued to better undertand how thee deciion-making role manifet themelve in activity deciion-making. Keyword Travel behavior Social network Deciion making Detination choice Introduction Current practice within travel demand rely on the ue of activity baed modeling method. Foundational to thi modeling framework i the concept of travel being a derived demand K. Deutch (&) K. G. Goulia GeoTran Lab, Department of Geography, Univerity of California, Santa Barbara, CA 93106, USA e-mail: deutch@geog.ucb.edu K. G. Goulia e-mail: goulia@geog.ucb.edu
756 Tranportation (2013) 40:755 771 from the neceity or deire to participate in activitie (but alo travel a a deirable activity per e). Thi paradigm ha rehaped the approach that i taken to modeling individual in a tranportation etting. It i being recognized however, that the aumption and implitic repreentation of activitie a being economically and pychologically driven i not all that i needed. Activitie many time are ocial in nature, and hould be modeled a uch. Even when activitie are not ocial in nature, it i poible that they are influenced by other ocial activitie that could contrain the time and pace dimenion of an activity (Páez and Scott 2007). Several attribute of activitie are conidered in modeling behavior a well a important factor that influence the choice proce. Although the literature i jut recently gaining momentum within travel behavior, the acknowledgement of the influence of other on time ue and travel behavior ha long been realized. For intance, Salomon (1985) made the claim that the deire for a ene of belongingne drive people to want to participate in activitie. Thi in turn drive the need for travel, a already dicued a a premie of the activity baed approach. In addition to thi, the time geography concept of coupling contraint ha been empirically examined by reearching the influence of ocial contact on an individual travel (Páez and Scott 2007). The broader concept of ocial network ha alo been explored by everal other (Axhauen 2005, 2007; Arentze and Timmerman 2008; Carraco and Miller 2006; Habib and Carraco 2011). A tated by Páez and Scott (2007), the need for ocial contact, and the effect of ocial influence on travel behavior, i one uch apect of deciion-making that deerve attention. Prior to thee exploration ariing in the mid to late 2000, other conideration in ocial influence had debuted in the travel behavior reearch community (Kitamura 1998). Detail uch a with whom activitie and travel were conducted (Harvey and Taylor 2000; Habib et al. 2008), or for whom the activity wa conducted (Goulia and Kim 2004) have made their way into urvey a intereting and thought provoking data type, leading to pioneering analye. Although the ocial apect of thee example are more broadly cat, reearch focued on undertanding within houehold interaction and the implication of thee interaction on time ue and travel behavior ha received mot of the attention regarding ocial influence (Gliebe and Koppleman 2002; Golob and McNally 1997; Yoon and Goulia 2010), and can be more eaily analyzed with a houehold level data collection exercie. In addition to the reearch dicuion baed on ocial influence, everal reearcher have focued more pecifically on the compoition of ocial network from a traditional ocial network definition. The original ue of the metaphor of a network to decribe a peron ocial relationhip came from a group of ociologit in Germany (Scott 1988). Social network are made up of node (people), which are connected by link. The analyi of thee ocial network, uing technique uch a graph theory, give reearcher a computational repreentation of the relationhip and poibly the connectivity between people (cloene, interconnectedne, etc.). Carraco and Miller (2009) break down everal characteritic into variou element. Firt, the compoition of a ocial network identifie the number of imilar relationhip to the individual (e.g., family, friend, coworker, choolmate, fellow church goer) and the level of cloene of each of thee type of relationhip. Second, they identify everal key characteritic defining the network tructure (the ize, intance of iolate or people only connected to the individual, denity, network ubgroup and potential of activity propagation from different type of relationhip or people). Thee element provide a theoretical bai for the development of urvey quetion ued in thi reearch to undertand ocial network and their contribution to deciion making. In order to develop the mot accurate model of deciion making and behavior, it i therefore important to keep in mind that the manner in which ocial network influence
Tranportation (2013) 40:755 771 757 behavior, and explore way in which they can be introduced into model. Much of the current practice in travel demand modeling rely on modeling the choice proce. It i recognized through everal theorie that ocial influence impact behavior, and therefore implicitly the deciion making proce. For intance the Theory of Planned Behavior (Ajzen 1991) include the influence of ocial norm. In addition, many theorie focu on the attainment of ocial capital, which include by nature ocial interaction and influenced deciion procee (Bourdieu 1984, 1998). Within travel behavior, reearcher have focued on everal apect of ocial activitie. For intance: telecommuting (Páez and Scott 2007), the propenity to conduct ocial activitie (Carraco and Miller 2006) and activity duration (Habib and Carraco 2011). In addition, reearch ha extended into examining who activitie are conducted with and their ocial nature (Sener and Bhat 2007), a well a both with whom and for whom an activity i conducted (Goulia and Kim 2004; Goulia and Henon 2006). We enviion developing choice model that explicitly incorporate the power in deciion making of individual in ocial network. Thee model will mot likely be tak and time allocation model with the important addition of repreenting power in a ytem that ha explicit unequal power among agent-role. Before developing the functional form and deriving the mathematical apparatu to etimate model of thi type we need to undertand the role played by individual in different deciion context. One example of thi negotiation and tak allocation within a houehold i the generation and allocation of ecort reponibilitie in a houehold (e.g., taking children to chool or a houehold member needing medical attention to the doctor) and it aociated houehold car type allocation (Bhat et al. 2012). In activity location choice or detination choice we do not have model that explicitly aign deciion role among the peron participating in the activity at the detination. To develop thi type of choice model in activity-baed model ytem it i very important to identify the power tructure in deciion making when group of individual participate in activitie. Thee concept have yet to be woven ufficiently into the framework of dicrete choice model, which are perhap the mot widely accepted model for deciion making. In order to do thi, we mut firt examine the role that different ocial network play in deciion procee, and determine how bet to repreent heterogeneity among ocial interaction. Data decription The data ued in thi tudy i a portion of a urvey conducted in Santa Barbara, California. The data collection conited of a mail recruit letter, with a web baed repone. The urvey included quetion about ocial network involvement, ize, trength and frequency of contact of the ocial network, and the role the repondent play in deciion making for activitie conducted with that pecific network type. The urvey alo included a ection of houehold and individual level ocioeconomic and demographic quetion, a well a everal additional ection regarding general deciion making linked to detination choice. The reulting ample tatitic are provided in Table 1 from a total of 574 repondent. Each repondent wa aked to elect from a lit of even different ocial network type the group in which they interacted with in a typical week. The lit of ocial network type wa developed uing reearch conducted by Carraco and Miller (2006) and Goulia and Kim (2004). Thi lit included immediate family, extended family, friend, coworker, tudent (peer), tudent (a a mentor) and organization member (religiou, port, club, etc.). Following the election of network, four quetion were aked for each of the ocial
758 Tranportation (2013) 40:755 771 Table 1 Sample decriptive tatitic Variable Decription Gender Female 59 % Male 41 % Employment Employed full time 44 % Employed part time 14 % Student full time 6 % Student part time 1 % Self employed 7 % Home dutie 4 % Unemployed 4 % Looking for work 1 % Retired 17 % Diabled 2 % Marital tatu Single, never married 23 % Married/dometic partner 61 % Other 16 % Relation to houehold Live alone 13 % Live with immediate family 72 % Live with extended family 3 % Live with friend 5 % Live with acquaintance 2 % Live with ignificant other 3 % Other 2 % Age Mean 49 year Houehold income Median $60,000 $69,999 Number of children Mean 0.47 Number of houehold member Mean 2.6 network elected regarding ize, trength, frequency of contact and deciion-making role. Figure 1 provide the quetion from the urvey. Method In order to undertand the way in which people are involved in different ocial network, and the role that they play in the deciion involved in thee group, latent cla cluter analyi wa ued. Latent cluter or group developed from the tatitical procedure were ued to firt claify apect of ocial network and their compoition, and econd undertand ocial interaction role. Latent Cla Cluter Analyi (LCCA) i a modeling technique within the latent cla model in which probabilitic method are employed to cluter or group object (or in our cae individual) into clae. Although the baic form of the LC cluter model i one with
Tranportation (2013) 40:755 771 759 Fig. 1 Page one and two of ocial network urvey quetion continuou indicator, extenion have been developed to accommodate mixed indicator type (including nominal and ordinal) and covariate to be imultaneouly modeled. The equation ued for LCCA with mixed indicator type i provided in Eq. 1. f ðy i jhþ ¼ XK Y J p k f k ðy ij jh jk Þ k¼1 j¼1 ð1þ
760 Tranportation (2013) 40:755 771 where y i i the peron repone (i = 1,,N) to the meaured variable and y i h i the ditribution of y given the model parameter h; N i the number of repondent; K i the number of cluter (k = 1,,K); p k i the prior probability of belonging to a latent cla or cluter k; J i the total number of indicator And y ij i each element of y i ued to individually pecify each univariate ditribution. Thee are the core for each repondent anwer of the quetion in Fig. 1. In addition to thi pecification, covariate can be ued to predict cla memberhip. When pecifying thee covariate, it i important to eparate them a exogenou variable ued only to predict memberhip, and not a endogenou variable ued to inform the development of cluter. Equation 2 provide the formulation for the incluion of thee covariate. f ðy i jz i ; hþ ¼ XK Y J p kjzi f k ðy ij jz i ; h jk Þ k¼1 j¼1 ð2þ where z i i the vector of the value of the covariate for individual i. In thi model pecification, the covariate are pecified a having direct effect, avoiding the influence of the covariate effect on the cla memberhip only going through the latent variable. The analyi wa conducted uing Latent Gold 4.5. To etimate the parameter, Maximum Likelihood (ML) and Poterior Mode (PM) method are traditionally ued. PM method account for the ue of everal prior (Dirichlet and Gamma) employed to avoid boundary olution or non-exitence of Maximum Likelihood etimate (Vermunt and Magidon 2005). In order to converge to a olution, Latent Gold etimation procedure include a two-tep ue of algorithm, firt uing Expectation Maximization (EM) and turning to Newton Raphon (NR) once a olution i near the Likelihood maximum. Model of different cluter tructure were etimated iteratively and compared. Model parimony, fit tatitic and cluter tructure were all ued to determine the appropriate number of cluter bet decribing the data and latent phenomenon. Conceptual framework In order to undertand both the compoition of different ocial network type and the different role that people have in thoe network, a two tage cluter model wa developed. The firt tep conited of developing a claification of intance of repondent ocial network involvement dependent on network compoition. In the econd tep, thee claification of ocial network involvement type were ued with deciion making repone to undertand difference in ocializer type, or the role people play in different intance of ocial network interaction. Social network compoition The analyi of ocial network compoition included three meaured attribute of the ocial network. The tated ize of the ocial network, perceived trength of the relationhip the repondent had with individual in the pecific network, and frequency of interaction (ee Fig. 1) with the ocial network were ued to create cluter of ocial network compoition type. Covariate of the type of ocial network were included to further drive the
Tranportation (2013) 40:755 771 761 etimation of cluter and claification. The conceptual model for thi tage of etimation i labeled a Model 1 in Fig. 2. Development of thi cluter model provided one claification for a number of ocial network attribute, which decribe a pecific intance of ocial interaction type. Each ocial group for each individual wa aigned a cluter cla a a reult of thi firt tage. Social engagement type Following model one, the cluter memberhip were ued to provide further inight into ocial apect and role. Model one claification were ued in combination with repone about the deciion making role (who decide the location where activitie take place) to develop ocializer type cluter (repreented a Model 2 in Fig. 2). Thee ocializer cluter were again clae of pecific intance of ocial network interaction for each repondent. Development of thee cluter wa ued to invetigate the poibility that difference in role exit among different ocial group type. Analyi In accordance with the conceptual framework provided in the previou ection, two latent cla cluter model were developed. The ample conited of 1,764 different intance of ocial network involvement from 574 repondent. Decription of the ocial network data i provided in Fig. 3. Repondent recorded participation on average with three different type of ocial network, with 98 % of repondent falling between one and five different ocial network type. Fig. 2 Conceptual model
762 Tranportation (2013) 40:755 771 Cluter model 1 (ocial network type) An iterative procedure wa ued to develop a erie of cluter model baed on ocial network apect provided both a exogenou and endogenou variable. Social network ize, trength and frequency of interaction were ued to inform the development of the latent cluter, while the type of network were ued a binary covariate. For etimation purpoe, one binary indicator (in thi cae organization) mut be left out of the model pecification. Each intance of ocial network involvement wa treated a an individual object to be claified in the cluter model, therefore claifying intance of participation. It i therefore poible for mot individual to have memberhip in different cluter, Fig. 3 Sample ocial network tatitic (N = 574, Mean: 3.07, Standard deviation: 1.231)
Tranportation (2013) 40:755 771 763 dependent on the ocial network involvement. The reulting model, a 5 cluter model wa determined to be the bet model repreenting the data baed on fit tatitic (provided in Table 2), model parimony and cluter tructure. The reulting profile of thi five-cluter model i provided in Table 2, and the probability mean are reported in Table 3. The five cluter developed are interpreted a hown in Fig. 4 that how the relative value of three criteria variable (network ize, trength of relationhip, and contact frequency) within each cluter. The covariate included in the model etimation provide inight into the type of ocial network that are preent in each cluter. Cluter one for example conit mainly of immediate and extended family, a well a friend. The probability mean indicate that intance of both extended family and friend have high probability of belonging to cluter one. Thi finding indicate that there i imilarity among thee three type of ocial network in the compoition of ize, trength and frequency, epecially in the cae of extended family and friend. Cluter two i largely repreented by immediate family ocial network intance. Thi cluter alo include a portion of the extended family and friend ocial network intance, but i motly dominated by immediate family. Thi reult i to be expected, a it how that network intance of immediate familie have qualitie of their compoition (relationhip trength, ize and level of interaction) that are not a common to other network type. Cluter three, four and five are primarily compoed of non-family or friend baed ocial network type. Commonality i again noticed, thi time between coworker ocial network type and tudent (either a mentor or peer) within both cluter three and cluter five. To further the explanation of cluter claification and ocial network type, a viualization of a cro-tabulation of cluter cla and network type i provided in Fig. 5. Notably, thi graph illutrate the trong domination of organization ocial network in cluter four. Cluter four primarily conit of large ocial network, with trength of relationhip in the middle to omewhat trong region on the pectrum. Cluter 3 appear to be dominated by profeional colleague and coworker/tudent. Cluter model 2 (deciion role) Following the development of a cluter model baed on ocial network type and attribute, the deciion role of individual with repect to the ocial network involved in were crotabulated. The reult of thi cro-tabulation were ued to examine the commonalitie and ditribution of deciion making role when deciding where activitie take place with other from a ocial network or role acro ocial network type. Deciion type were categorized into five group a a reult of repone from the urvey. The firt three deciion type correpond to each of the repone option of the urvey, which have been hortened for eae of reference. Repone of I generally have a large ay in the deciion making proce were termed leading deciion maker, I partake in deciion making, but not more than mot other were termed equal collaborator, and I uually jut go along with deciion made by other were termed deciion follower. Additionally, the urvey form allowed for an other repone, allowing repondent to explain their election of other. Many of thee explanation indicated the fixed nature of activitie with thee ocial group. For intance explanation like uually fixed meeting place or The location of volunteer activitie I participate in i already known were given. Individual electing the other option for their role were categorized a other deciion-making role. Latly, due to the fact that repondent were allowed to elect multiple repone variable decribing their deciion making role in the network, a fifth categorization wa created. The explanation repondent gave for electing multiple role conited of tatement uch
764 Tranportation (2013) 40:755 771 Table 2 Model One Profile Indicator Covariate ize trength contact immediate extended friend coworker peer mentor Cluter1 Cluter2 Cluter3 Cluter4 Cluter5 Cluter Size 0.3815 0.2209 0.1604 0.1544 0.0827 1-5 peron 0.5074 0.9441 0.4452 0.0748 0.2698 6-10 peron 0.2907 0.0533 0.2976 0.1376 0.2733 11-20 peron 0.1421 0.0026 0.1697 0.2160 0.2362 21-50 peron 0.0474 0.0001 0.0661 0.2314 0.1393 51-100 peron 0.0083 0.0000 0.0135 0.1297 0.0430 Over 100 peron 0.0042 0.0000 0.0079 0.2104 0.0384 Mean 1.7709 1.0586 1.9287 3.8350 2.5274 1 0.0001 0.0000 0.0210 0.0094 0.0030 2 0.0003 0.0000 0.0346 0.0181 0.0072 3 0.0022 0.0000 0.0920 0.0568 0.0275 4 0.0056 0.0000 0.0941 0.0684 0.0407 5 0.0334 0.0001 0.2230 0.1909 0.1392 6 0.0720 0.0009 0.1917 0.1934 0.1727 7 0.1540 0.0081 0.1635 0.1943 0.2126 8 0.3088 0.0652 0.1308 0.1831 0.2456 9 0.2050 0.1746 0.0346 0.0571 0.0939 10 0.2185 0.7511 0.0147 0.0286 0.0576 Mean 8.2077 9.6665 5.6571 6.2290 6.8689 Everyday 0.0255 0.7264 0.0014 0.0178 0.4359 A few time a week 0.3873 0.2689 0.0617 0.3213 0.5317 Once a week 0.2744 0.0046 0.1254 0.2702 0.0303 A few time a month 0.2316 0.0001 0.3039 0.2707 0.0021 Once a month 0.0524 0.0000 0.1974 0.0727 0.0000 Le than once a month 0.0287 0.0000 0.3102 0.0473 0.0000 Mean 2.9843 1.2785 4.5649 3.2010 1.5986 0 0.8462 0.1011 0.9889 1.0000 0.9919 1 0.1538 0.8989 0.0111 0.0000 0.0081 Mean 0.1538 0.8989 0.0111 0.0000 0.0081 0 0.8605 0.9596 0.9661 0.9974 0.9998 1 0.1395 0.0404 0.0339 0.0026 0.0002 Mean 0.1395 0.0404 0.0339 0.0026 0.0002 0 0.3807 0.9672 0.9138 0.9961 0.9388 1 0.6193 0.0328 0.0862 0.0039 0.0612 Mean 0.6193 0.0328 0.0862 0.0039 0.0612 0 0.9335 0.9894 0.3590 0.9934 0.4316 1 0.0665 0.0106 0.6410 0.0066 0.5684 Mean 0.0665 0.0106 0.6410 0.0066 0.5684 0 0.9942 0.9999 0.8874 0.9157 0.9367 1 0.0058 0.0001 0.1126 0.0843 0.0633 Mean 0.0058 0.0001 0.1126 0.0843 0.0633 0 0.9975 1.0000 0.9083 0.9095 0.7072 1 0.0025 0.0000 0.0917 0.0905 0.2928 Mean 0.0025 0.0000 0.0917 0.0905 0.2928 BIC = 15529.1694, Claification error = 0.1091 a it depend on the deciion or there are three of u, and we all at time feel what one want to do i more important than other. Thee multiple repone intance were collaped into one variable, and were categorized a mixed deciion role. Reult of the cro tabulation are provided in Fig. 6. Of note, deciion follower primarily manifet within colleague/coworker ocial network, and organization intance. The equal deciion making role i repreented in each of the network type,
Tranportation (2013) 40:755 771 765 Table 3 Model 1 probability mean Indicator Covariate Cluter1 Cluter2 Cluter3 Cluter4 Cluter5 Overall 0.3815 0.2209 0.1604 0.1544 0.0827 ize 1-5 peron 0.3669 0.4112 0.1497 0.0205 0.0518 6-10 peron 0.5623 0.0543 0.1991 0.0956 0.0888 11-20 peron 0.4166 0.0045 0.1821 0.2749 0.1220 21-50 peron 0.1980 0.0000 0.1698 0.4685 0.1638 51-100 peron 0.0793 0.0000 0.1126 0.6188 0.1893 Over 100 peron 0.0359 0.0000 0.0284 0.8564 0.0793 trength 1 0.0206 0.0000 0.4998 0.2866 0.1929 2 0.0063 0.0000 0.6069 0.2925 0.0942 3 0.0278 0.0000 0.5250 0.3183 0.1289 4 0.0336 0.0000 0.4233 0.4588 0.0843 5 0.1590 0.0002 0.4144 0.3012 0.1253 6 0.2487 0.0001 0.3647 0.2772 0.1094 7 0.4720 0.0120 0.1813 0.2104 0.1242 8 0.5681 0.0703 0.1013 0.1558 0.1044 9 0.5625 0.2887 0.0262 0.0474 0.0751 10 0.3208 0.6324 0.0084 0.0223 0.0161 contact Everyday 0.0438 0.7715 0.0012 0.0105 0.1730 A few time a week 0.4877 0.1855 0.0343 0.1513 0.1412 Once a week 0.5547 0.0105 0.1180 0.3022 0.0146 A few time a month 0.5496 0.0003 0.2577 0.1914 0.0010 Once a month 0.2980 0.0000 0.5350 0.1669 0.0000 Le than once a month 0.1370 0.0000 0.7298 0.1333 0.0000 Immediate 0 0.4361 0.0302 0.2143 0.2086 0.1108 1 0.2259 0.7646 0.0069 0.0000 0.0026 Extended 0 0.3523 0.2274 0.1663 0.1653 0.0888 1 0.7826 0.1312 0.0799 0.0059 0.0003 friend 0 0.1971 0.2899 0.1989 0.2087 0.1054 1 0.8984 0.0275 0.0526 0.0023 0.0192 coworker 0 0.4336 0.2660 0.0701 0.1868 0.0435 1 0.1421 0.0131 0.5758 0.0057 0.2633 peer 0 0.3945 0.2297 0.1481 0.1471 0.0806 1 0.0572 0.0004 0.4688 0.3379 0.1358 mentor 0 0.4023 0.2334 0.1540 0.1485 0.0618 1 0.0174 0.0002 0.2732 0.2595 0.4497 although i mall in the cae of interaction with tudent a a mentor. Thi ocial network type i predominately compried of leading deciion maker who have the mot influence in the deciion, which i an intuitive role of omeone in a mentoring relationhip. In addition to the cro-tabulation, a econd latent cla cluter analyi wa conducted to examine the tated role repondent have in deciion-making procee among the cluter developed by network attribute. The memberhip claification of the latent cla cluter model previouly dicued were ued a an indicator in the etimation of the econd model. In addition to claification reult, the tated deciion-making role variable wa ued in the development of cluter. An iterative procedure wa again ued in pecifying the model tructure. The fit tatitic (provided in Table 4), cluter tructure and claification error were ued to guide the final acceptance of the four-cluter model. Reult of thi econd model are provided in Table 4 (profile) and Table 5 (probability mean). The reult of thi cluter analyi provide ome intereting inight on deciionmaking tyle within different ocial context. Cluter can be decribed a:
766 Tranportation (2013) 40:755 771 Cluter one: mall network ize trong relationhip trength weekly frequency of meeting Cluter two: very mall network ize very trong relationhip trength daily frequency of meeting Cluter three: mall network ize medium trength of relationhip monthly contact Cluter four: large network ize medium trength of relationhip weekly contact Cluter five: mall to medium network ize medium to trong relationhip daily contact w d w d w d w d w d l m l m l m l m l m ize trength frequency ize trength frequency ize trength frequency ize trength frequency ize trength frequency Fig. 4 Cluter reult of network attribute (note that for ize mall, l large; for trength w weak, trong; and for frequency d daily, m monthly) Cluter one Thi cluter i largely compried of family and friend with mall ize, trong relationhip and frequent interaction. The predominant deciion role in thi cluter i either leading deciion maker having large influence, or equal collaborator in the deciion. Cluter two Thi cluter i mainly compried of non-family or friend ocial network, and it i the cluter with the highet probability for organization intance. The deciion-making role i motly deciion follower, with ome equality of deciion making with a collaborator. Cluter three Thi cluter i compried motly of the mall ize, trong relationhip, weekly interaction cluter, which i largely baed on ocial network of friend. The deciion making role for thi cluter i motly deciion follower or mixed deciion trategie. Cluter four Thi cluter i compried motly of organization, mentor and coworker ocial network type that are mall, and medium relationhip trength and everyday interaction. The deciion-making role for thi cluter i compried primarily of leading deciion maker, and ome intance of equal collaboration. Interetingly, cluter one and three exhibit many imilaritie in the compoition of ocial network type, a do cluter two and four. The bifurcation of thee cluter group occur
Tranportation (2013) 40:755 771 767 Fig. 5 Social network cluter memberhip by type Fig. 6 Deciion making type by ocial network type
768 Tranportation (2013) 40:755 771 Table 4 Model 2 profile Cluter1 Cluter2 Cluter3 Cluter4 Cluter ize 0.5227 0.2395 0.7 0.1141 Indicator Model 1 cluter claification 1 0.5523 0.0000 0.8007 0.0000 2 0.3663 0.0044 0.1850 0.0001 3 0.0811 0.4045 0.0143 0.0655 4 0.0002 0.4701 0.0000 0.3982 5 0.0000 0.1209 0.0000 0.5362 Mean 1.5293 3.7075 1.2136 4.4704 Deciion type Leading deciion maker 0.3537 0.0002 0.0001 0.6250 Equal collaborator 0.6438 0.2355 0.1 0.3745 Deciion follower 0.0025 0.5444 0.4474 0.0005 Other deciion tyle 0.0000 0.0943 0.1336 0.0000 Mixed deciion tyle 0.0000 0.1256 0.3067 0.0000 Mean 1.6488 3.1097 3.6345 1.3754 BIC = 9783.132, Claification error = 0.0717 Table 5 Model 2 Probability Mean Cluter1 Cluter2 Cluter3 Cluter4 Overall 0.5227 0.2395 0.7 0.1141 Indicator Model 1 cluter claification 1 0.7442 0.0000 0.2558 0.0000 2 0.8908 0.0032 0.1059 0.0001 3 0.2838 0.6525 0.0116 0.0521 4 0.0007 0.7175 0.0000 0.2818 5 0.0000 0.3152 0.0000 0.6848 Deciion type Leading deciion maker 0.7210 0.0002 0.0000 0.2788 Equal collaborator 0.7494 0.1217 0.0341 0.0948 Deciion follower 0.0058 0.6865 0.3074 0.0003 Other deciion tyle 0.0000 0.8028 0.1971 0.0000 Mixed deciion tyle 0.0000 0.3658 0.6342 0.0000 due to the difference in role of deciion-making, with deciion follower being clearly repreented in cluter two and three. The manifetation of thee different ocial role within two et of imilar cluter compoition indicate that there are both difference in deciion-making role acro different type of ocial network, a well a heterogeneity within imilar ocial network type. Concluion It i widely accepted that the involvement in activitie in different place i a driver of the need for travel. Often time, thee activitie have a ocial component, which influence
Tranportation (2013) 40:755 771 769 either when or where the activity occur. Many of thee ocial interaction are difficult to capture in current urvey methodologie. Houehold interaction are the exception to thi tatement, becaue many travel behavior urvey are collected at the unit of houehold level. A intrahouehold interaction become an important component of explaining travel behavior, it i important to realize that imilar influence occur outide of the houehold unit. The ability to more accurately predict not only the patial, but alo the temporal attribute of an activity depend on the incluion of important information. Although thi reearch focue primarily on the detination choice proce, it i important to note that a further need and reearch direction i the expanion of thi deciion making analyi to additional attribute of activitie uch a temporal (daily activity agenda and cheduling of pecific activitie) or even the overall ocial compoition of the activity (ize, ocial network type, etc.) and how thee influence future activitie. To undertand the role of different ocial network in the live of individual, we mut firt undertand how they differ from each other. A latent cla cluter analyi wa conducted to examine difference and imilaritie among different ocial network type, with repect to the ize, trength of relationhip and the frequency of interaction. Reult how imilaritie with thee attribute among family (immediate and extended) and friend, a well a organization, coworker/colleague, tudent (a both peer and mentor). In addition to finding imilaritie, difference tood out a well. For intance, many of the very trong, mall family relationhip were preerved in a pecific cluter. In addition to the difference and imilaritie of network compoition and type, the deciion-making proce among thee ocial network exhibit imilar trend. The deciion-making role of an individual can differ vatly acro different ocial engagement type. For intance, a parent ha a much different role a a member of a family for which he or he i the head; veru the role he or he play a a member of a company, or friend. The reult of the econd cluter analyi revealed different group of deciion-making trategie within imilar ocial network type, a well a imilaritie in deciion making trategie acro different ocial network type. Thi i particularly important for all facet of activity and travel behavior model that aim at decribing the deciion proce followed by individual and their group. The reearch here how we can identify deciion-making role (leader v. follower) and context (family v. friend ocial network). It i alo poible thee role change with the type of activity or other circumtance. Knowing all thi will increae our ability to predict where people will go to participate in activitie and alo who hould be influenced to motivate a group of people in adapting behavior that are aligned with policie (e.g., utainability). In addition to ocial influence to behavior adoption, the invetigation of ocial network can provide inight into the patial ditribution of joint activitie. An important next tep of thi reearch i to determine the pattern of detination choice with repect to the location of individual prior to a joint meeting. Future data collection and analyi will involve examining activity diarie of individual and exploring the convergence of time pace prim of member of different ocial network type in detination choice. Thi will allow for invetigation a to whether there i correlation between the proximity (cloer, equiditant or further) of detination to a pecific individual and the deciion-making role. It i quite poible that detination choice for joint activitie have a patial bia toward a more vocal deciion maker due to the cognitive proceing of alternative and mental map repreentation of pace. Thi however mut be explored empirically, and require unique data for the invetigation. An enhanced undertanding of the proce of deciion making in thi vein a well a a more general knowledge of the joint deciion making proce will no doubt enhance current modeling effort. In addition, increaing our undertanding of ocial
770 Tranportation (2013) 40:755 771 behavior will provide a richer theoretical bai for the aumption implicit in the activity baed modeling paradigm. Thi reearch wa focued pecifically on the ocial network compoition and deciion making trategie apparent in different network. Of equal importance however i an undertanding of the individual and hi or her memberhip in different ocial network a well a deciion-making type. Future work include conducting a peron-baed analyi, imilar to the one preented in thi paper, to determine whether it i feaible to predict or model ocial engagement type with repect to known ocio-demographic indicator and memberhip in different life cycle tage. In addition, thi data will be combined with a econd phae of data collection coniting of an activity diary and martphone baed activity log. Deciion making procee occurring for pecific oberved activitie will be compared to the ocial engagement type and role provided by the individual during the firt phae of the data collection. Acknowledgment Funding for thi project wa provided by the Univerity of California Tranportation Center, the United State Department of Tranportation Eienhower Fellowhip program, the Univerity of California Office of the Preident UC Lab Fee Program on Next generation Agent-baed Modeling and Simulation, the Multicampu Reearch Program Initiative for Sutainable Tranportation, and the Univerity of California Preident Diertation Fellowhip. Reference Ajzen, I.: The theory of planned behavior. Organ. Behav. Hum. Deci. Proce. 50(2), 179 211 (1991) Arentze, T., Timmerman, H.: Social network, ocial interaction and activity-travel behavior: a framework for microimulation. Environ. Plan. Part B 35, 1012 1027 (2008) Axhauen, K.: Social network and travel: ome hypothee. In: Donaghy, K., Poppelreuter, S., Rudinger, G. (ed.) Social apect of utainable tranport: tranatlantic perpective, pp. 90 108. Ahgate, Alterhot, Hant (2005) Axhauen, K.: Activity pace, biographie, ocial network and their welfare gain and externalitie: ome hypothee and empirical reult. Mobilitie 2(1), 15 36 (2007) Bhat, C. R., Paleti, R., Pendyala, R. M., Goulia, K. G.: SimAGENT activity-baed travel demand analyi: framework, behavioral model, and application reult. SimAGENT Core Model. Phae 2 Final Report 4 Submitted to SCAG, Santa Barbara (2012) Bourdieu, P.: Ditinction: a ocial critique of the judgment of tate. Harvard Univerity Pre, Cambridge (1984) Bourdieu, P.: Practical reaon: on the theory of action. Stanford Pre, Stanford (1998) Carraco, J., Miller, E.: Exploring the propenity to perform ocial activitie: a ocial network approach. Tranportation 33, 463 480 (2006) Carraco, J., Miller, E.: The ocial dimenion in action: a multilevel, peronal network model of ocial activity frequency between individual. Tranp. Re. Part A 43(1), 90 104 (2009) Gliebe, J., Koppleman, F.: A model of joint activity participation between houehold member. Tranportation 29(1), 49 72 (2002) Golob, T., McNally, M.: A model of activity participation and travel interaction between houehold head. Tranp. Re. Part B 31(3), 177 194 (1997) Goulia, K.G., Henon, K.M.: On altruit and egoit in activity participation and travel: who are they and do they live together? Tranportation 33(5), 447 462 (2006) Goulia, K.G., Kim, T.: An analyi of activity type claification and iue related to the with whom and for whom quetion of an activity diary. In: Timmerman, H. (ed.) Progre in activity-baed analyi. Chapter 14, pp. 309 334. Elevier, Amterdam (2004) Habib, K. and Carraco, J.: Invetigating the role of ocial network in tart time and duration of activitie: a trivariate imultaneou econometric model. Preented at the Tranportation Reearch Board meeting, 2011 Habib, K., Carraco, J., Miller, E.: Social context of activity cheduling. Dicrete-continuou model of relationhip between with whom and epiode tart time and duration. Tranp. Re. Rec. 2076, 81 87 (2008)
Tranportation (2013) 40:755 771 771 Harvey, A., Taylor, M.: Activity etting and travel behavior: a ocial contact perpective. Tranportation 27(1), 53 73 (2000) Kitamura, R.: An evaluation of activity-baed travel analyi. Tranportation 15, 9 34 (1998) Páez, A., Scott, D.: Social influence on travel behavior: a imulation example of the deciion to telecommute. Environ. Plan. 39, 647 665 (2007) Salomon, I.: Telecommunication and travel: ubtitution or modified mobility? J. Tranp. Econ. Policy. 19(3), 219 235 (1985) Scott, J.: Social network analyi. Sociology 22, 109 127 (1988) Sener, I.N., Bhat, C.R.: An analyi of the ocial context of children weekend dicretionary activity participation. Tranportation 34(6), 697 721 (2007) Vermunt, J.K., Magidon, J.: Technical guide for latent gold 4.0: baic and advanced. Statitical Innovation Inc., Belmont (2005) Yoon, S., Goulia, K.: Contraint-baed aement of intra-houehold bargaining on time allocation to activitie and travel uing individual acceibility meaure. Preented at the Annual Meeting of the Tranportation Reearch Board, 2010 Author Biographie Kathleen E. Deutch i in her lat year a a doctoral tudent in the Department of Geography at UC Santa Barbara. Her main reearch interet include activity baed modeling and data collection. Her diertation work focue on the modeling of detination choice incorporating place baed attitude and ocial network component. Kontadino G. Goulia i a profeor of tranportation at the UC Santa Barbara Department of Geography. He erved a profeor in Civil and Environmental Engineering at PennState Univerity from 1991 to 2004, chaired both the Tranportation Reearch Board (TRB) Tak Force on Moving Activity Baed Approache to Practice and Traveler Behavior and Value committee. Goulia edited two book and publihed more than 230 reearch report and paper and i the co-founder and co-editor in-chief of the journal Tranportation Letter.