The Mediation Model of Influence ofWeb site Performance on Needs for Crisis Communication in the Case ofToxic Milk Powder

Most lireraturcs concerning Internet public relat ions (PR) assumcd rhat "the viewpoint of traditional PR could be applied ro Internet PR." Howcver, rhis assumption lacks empirical evidence and validiry. This a nicle is a case study on rhe crisis of toxic milk powder in 2008, for testing r:his assumption, a n online ex periment (N=402) was conducted. Resea rd1 findings suggested that the effect of corporare web si re performance on the needs for crisis communi cation was mediared by a ser of mediarors (i. c., OPR, crisis responsibiliry a nd sclf-efficacy), with OPR bei ng rhe most cr itica! in the model. The impona nce of rclarionship managemenr in both tradicional and Internet PR is madc cvident as a resulr of r:his rcsearch.

The rise of rhe Internet makes great influence on co mm uni cation enviro nment, and therefore the com munication srrategies of organizations face rhe phase of rransforming (Sun, 2000). The Internet provides companies a platform for srrategic talking, interaction, and shaping longrerm relationship with its stakeholders (Kent & Taylor, 1998). Scholars had already regarded rhe Internet as one of crisis communication media, and so do their research approaches (e. g., Taylor & Kent, 2007;Perry, Taylor, & Doerfel, 2003), and sorne of rhem further examined the concept of tradicional theories from this approach.
These studies preserved wirh rheir own argument. Sorne focused on ei rher each web sire design (e. g., inreractivity) or media effect. However, rhose studies fell short on interprering all crisis phases. Sorne were limited ro en umerare principies for crisis communícation or future applicarion of che Internet but lacked rheoretical basis. Now rhe quesrion falls on whether viewpoint of tradiciona l PR applies ro Internet PR or not. Except for platform features (e. g., interacrivity, immediacy, anonymity and trans-border; Sun, 2000), what is rhe va l ue of web si tes for organizations? In addirion, li rtle is known abour ways organizaci ons' web sites integrare differenr cr. isis phases ro communicate, rhe way public percepc and the appreciable crireria on Internet crisis com munication.
Accordingly, chis article is based on crisis of coxic milk powder (China milk scandal) in 2008, intending co integrare cr isis phases to examine the relationship betwee n web sice design and c risis communicar ion, and how do public percepc it.
More specifically, ch e m ain purpose of chis arricle is nor ro preve the excell ent effect of organizati ons' web sires on building relarionship and crisis communicar ion, but to find out that if rhe perspective of tradicional PR co uld ex rend ro rhe Inrern er, and rhe way web site performance associated wirh crisis communicarion. Hence, currenr research focused on two questions: (1) What is the relationship berween web site design and crisis communication messages on corporate web sites? (2) What is the influencia! mechanism of web si re performance on needs for crisis communicarion? PR and Internet PR by putting the currenr condition of web si te performance and crisis communication on rhe Internet, and the perception of public togerher.
2. Context of Theory 2.1 . Crisis ofToxic Milk Powder in Taiwan Si nce Sep. 2008, rhe case of toxic mil k powder had gane up ro a crisis of global food safety from only a crisis of single milk powder m an ufacturer in China (the San lu group), arousing concern for people in Taiwan (should be nociced rhat rhe "crisis of toxic milk powder" which termed in this article only refer ro che follow-up effect chac spread to Taiwan).
This case could be divided into chree srages: (1) Pre-Crisis (from Mar. to Sep. 12th): In chis scage, none of media direcrly indicaced the brand name of che corporarion (r he Sanlu grou p). People in Taiwan didn't pe rceive the rhrea r of crisis; (2) Durin g the Crisis (from Sep. 13th ro mid-Occ.): There were massive news reports focusing on rhe crisis and responsibili ty amiburion, arousing a high degree of conce rn regarding to food safety in Taiwan in th is srage; (3) After the Crisis (after rh e late Oct.): Though rhe eve nc wasn't yet over, the public were no longer frightened. In addicion, media shifred their focus from respo nsibility acco unrin g to che mea ns of sequential managem ent.

The Internet a nd Crisis Communicarion
The sr raregic and cacrical purpose of crisis com munication us ually is reducin g che harm ro image of an organizacion; someci mes ir even makes a more posirive effect on organizatio n' s image compared to rhe image of che company befare crisis (Wu, 2000(Wu, , 2005. The Internet plays mu l tiple roles in the process of crisis cornm unica tion, and i t may noc on ly intensify rh e crisis, but also be a channel for communicarion. However, the vie wpoint of Intern et PR is in irs infancy and slighd y with a prorotype (Wu, 2005 entations, public relations approach underlines the overall strategies and principies of organizational crisis management, and rhetorical approach discusses the image manageme nt strategy organ izations adopted (e. g., Huang, 2006;Wu, 2000Wu, , 2005. Whatever approach was adopted, literatures mostly focused on the communication strategy betwee n organizations and its stakeholders in crisis (Uimer, 2001 ). And, from the perspective of crisis management, an organization should communicate to public continuously to minimize the potential crisis regardless of the stage of the crisis (Fearn-Banks, 2002;Taylor, Vasquez, & Doorley, 2003). The relations hip between organizations and public would affect the crisis responsibility which was attri buted toan organization (Benoit, 1997;Coombs & Holladay, 2001, 2002Park & Le n-Rios, 2004). The perspecti ve which extend to pre-crisis relationship reflects that the gradual merging of two orientations (Witkemper & Pri tchard, 2007), and points out the importance of relationship (e. g., Ulmer, 2001;Wu, 2000). Coombs & Holladay (1996) also em phasized the importance of relationship and incorpora ted the concept of attribution theory in their study. They proposed an experiment to examine how type of crisis, organ izational performance history and crisis response strategy associated with image of an organizarían in terms of attribution theory. They assumed that if organizational performance history is more negative (i. e., high frequencies of occurrence of similar crises), public will consideran organizarían having more responsibility for the crisis, resulting in more harmful negative influence. Research results su pported this hypothesis. Coombs (1998) took organi zatio nal performance history further into two parts: (1) crisis history: the same concept to organizational performance history in Coombs & Holladay (1996); (2) relation history: the pre-crisis relationship between organizations and public. Accordingly, Coombs & Holladay (2001) proposed a crisis situation model and indicated tha t crisis performance histo ry would affect assessment for responsibility and image. Relarive to crisis responsibility, the relationship between organizational performance history and assessment for im age is much closer, and the explanatory power of relation history is much stronger than crisis history. When the relationship between organizations and public is poor, public would evaluare the image more negatively, whi le neutral or good relationship makes no differences. If there we re no existing data about relation history and crisis history, public would tend to predict a more positive relationsh ip. Since Coombs & Holladay (1996, 2001 integra red variables of situation identification, relationship management, crisis responsibility a nd image of organizations in crisis com munication field, leading furrher research to identification and disc ussion for the relationship a mo ng those variables (e. g., Coombs & Holladay, 2002;Huang, 2006;Kim, 2008;Kang, Garciaruano, & Lin, 2008;Park & Len-Rios, 2008;Wu, 2005). Their model took the effect of relation history on crisis response strategy into account but lost sight of related variables (e. g., crisis communication strategy and practice; Wu, 2000). In fact, crisis response strategy could reduce the da mage on image of an organization by adjusting public pe rception. Thus, this model was amended soon after by adding a variable of organizarían reputation (Coombs & Holladay, 2002). Moreover, Kang et al. (2008) found that if people have low relationship sa tisfaction, they tend to take actions that ca used potential crisis (e. g., to tra nsmit the rumors). Even o n the Internet, the relationship between organizat ions and public is also a crucial interpretation variable to crisis communication. Sun (2004) defined Intern et PR as: "The featu res of th e Internet built up a platfor m for individuals or organizat ions to interact and communicate with each other. And features ca n narrow the gap between corporations and public or take them into the corporate promotion content (p.3)." Kent, Taylor & White (2002) termed these functions or features as "web site design". As a result, rhis article defined the relationship built through the Interne t as "the dialogic relationship between organizations and publíe which is based on the web si te design organizations presented, including the content of dialogue a nd techniques of promoting dialogue".
When discussing crisis communication on the Internet, the relationship built through the Inte rnet s houldn't be ignored. Without web site designs, organizations' si tes will lose the possibility ofbuilding a relation- ship wi th public. Since these web site designs are also the criteria for assessin g web site performance (Schneider, Frechtling, Edgar, Carwley, & Goldstein, 2000), its composite measure could be a variable for estimating global web si te performance.
The current study on ly took the influence of relation history (i. e., organ izat ion-public relationship) into account since the crisis of tox c mil k powder happened for the fi rst time. But how does rhe res ult of o rdtnary relationship built affect the effectiveness of crisis communication? Crisis responsibility seems ro provide a reasonable expl anation. When rhere is a crisis, public want to know the responsibility amibution (Fearn-Banks, 2002). If there was a better relationship between organizations and public, organizations may bear less responsibiliry. Note that crisis responsibi lity is a factor of subjective evaluation perceive d by public; ir doesn't matter whet her an organ izarían is actually responsible (Benoit, 1 997).
I n sum, the cu r rent research infers that the relationship between organizations and public and the crisis responsibility attrib ution wo uld affect public's needs for crisis comm unication messages and organizational crisis response strategy, thereby affecting eventual effectiveness of crisis communication.

The relationship between the Internet and crisis communication
Previous researches mainly adopted two approaches brought up by Taylor & Kent (2007) when ir comes to discussing the role of the Internet played: (1) The effectiveness of new media: This line of researches are case studies ofhow rhe Internet helps in crisis communication or conceptual studies of application and functional process of the Internet in organizational communication (e. g., Coom bs, 1998b); (2) The interaction between new and tradicional media: Another l ine of researches focus on how rhe Internet integrated or merged into tradicional media (e. g., Caldiero & Taylor, 2005;Taylor & Perr y, 2005;Perry et al., 2003). lt implies that the interaction of between co mmon foundation media integration should a lso be taken into accounting when ir comes to discussing the role as a new media of the I nternet.
Simil ar prerequisite could be found by reviewing the previous studies: Perspective of tradicional PR co uld also be applied to Internet PR. Kent & Taylor (1998) implied that "relationship bui lding is the foundation of public relations (p.324), " mentioning tha t scholars haven't examined the effectiveness of the Internet on bui ldi ng relationship between organ izations and public. Sun (2004) also defined Internet PR as a process of usi ng Internet features ro interact with public for establishing relationship. They seemed to assume tbat both Internet and trad itional PR have the same foundation, contributing to bui ld up a relationship. Furthermore, Taylor & Kent (2007) claimed thar "tradicional media relarions tactics can be adapted for use on the organization's Web site during and after a crisis (p. l 41)", seeing crisis comm unication on the Internet as simply a matter of the extension of tradicional crisis communicat ion.
Most resea rches on advantages in crisis communication and public relations were theoretical and ex ploratory studies, which focused main ly on furure implications. In addition, those studies placed their emphasis on the characterisrics of new media (e. g., Aspden & Katz, 2000;Taylor & Kent, 2007;Taylor, Kent, & White, 2001;Sun, 2004). The hypotheses were rarely be examined. Li trie is kno wn about whar the effecr public relations strategy having and how crisis communicarion implementing through the Internet, and what theoretical basis it basi n g on. Huang (2001b) suggested that che core value of tradicional PR is relationship. But is it also the same to Internet PR? lt's still an outstanding issue.
The study of Caldiero & Taylor (2005) had found more specific evidence supporting the effect of crisis communication tbrough the I nternet. They found that integrating rate of full quotes was 12%, noting that "if an organizarían has a better than one in 1O chance of getting a quote th at frames a n issue appear in a n ews story, tben ir makes sense to issue a news release in a crisis (p.1 6)." However, their study was still based on a framework ofBenoit's theory of image restor ation (which is also the viewpoint of tradicional public relat ions and crisis management).
Only if this ass umption (i. e., rel a tionsh ip building is the common foundation of tradicional PR and Internet PR) is true will the argument of previous studies have its foothold. The "public" which term ed in current study on ly refers to consumer, mainly based on two reasons: First, consumer is rhe primary public which organization's web sires communicate to (Esrock & Leichty, 2000;Sun, 2004), and thus subject (consumer) variables took an important role in crisis; Second, previous researches had discussed rhe roles orher public played in Internet crisis communication, such as media or interna!staff (Caldiero & Taylor, 2005;Hi ll & White, 2000;Perry et al., 2003), but less dealr with the perception of consumer for Internet crisis communicat ion, let al one invesrigated the quest ion rhat if crisis managemenr through the Internet affect the consumer or not.
In rhis part, this arricle adopted the definition of public relations of Grunig & Hunt (1984), attribution theory ofWeiner (1995) (as cited in Coombs, 1998a) and the concept of self-efficacy of Bandura (1982).

Web site performance
Generally speaking, the criteria to assess we bsi te are listed as follows, contenr, design and aesthetics, the disclosure of author, s ponsor and bui l der, immedi acy of information (includ i ng frequency of web si te updare and maintenance), copyright ofinformati on, ease for use, accessibility and availability (Schneider et al., 2000). According ro Schneider et al., web sire design is a crirerion for measuring web site performance, consistent with Kent et al. (2002), while they provided the theorcrical basis of criteria: rheory of dialogue.
The essence of public relarions is a dialogue, the conAicrion resolurion and negotiarion, not only the persuasion or media effect (Grunig & Hunt, 1984). The relarionsh i p between organizations and public can be de veloped and a cr isis can be processed through web site design. The ult i mare goal ofbui lding a si te is to develop a fine dialogical relationshi p (Kent & Taylor, 1998).
According ro previous category const ruction of we b si te design (e. g., Ha & James, 1998;Sun, 2004), most studies were based on five principies ofKent &Taylor (1998) listing as follows; A di alogue first i nvol ves attraction ofinteractions (i. e., useful ness of i nformarion); for de veloping relationship, must ro interacr (i. e., ease of interface); for growing rela-tionship, must to dialogue (i. e., conservarion of visitors); and for thriving rel ationship, must to occur maintenance and satisfactory interactions (i. e., generation of rerurn visirs and dialogic loops) (Taylor et al., 2001, p.268).
Except for these principies above-mentioned, credibility should be taken into consideration. Credibility is essentia l to the practice of public relations (Coombs & Holladay, 1996). Previous studies concerning CMC (computer-mediated communication) a nd interpersonal com munication on the Intern et have also found rhat all types of relationship involvi ng trust (Tayl or et al., 2001;Green, 2007;Riegelsberger, Sasse, & McCarthy, 2007).
Wu (2000) suggested that the relationship between organizations and public is crucial to crisis communicat ion . The views of Kem & Taylor (1998) on rhe web sire design also implied to rhe web site performance (which is also rhe composite measure of all web site designs) of an organization wo uld influen ce rhe relationship between organ izations and public.

Needs for crisis communication
The needs for crisis com munication representing "the characreristic of inform a rion which publ ic pay atten tion to during the crisis," were rhe factor of public perception. During a crisis, there are many ch aracreristics ofinformation which public valued, such as the usefulness or immediacy of crisis comm unicarion message. And these needs wou ld cha n ge when changing rhe percep tion of public.
When browsing an organ ization's si te, rhe public's needs for crisis communicarion would be affecred by web si re perform ance which public perceived during crisis because of th e mediation of organiza tion-public relationship (OPR), crisis responsibility and self-efficacy.

OPR
Grunig & Hunr (1984) defrned public relations as "a relarionship exisring between organizarions and its stal<eholders" (i. e., OPR). The val ue of public relations is the l ong-rerm relarionsh i p developed between organizations and publ i c, which is consistent wir h concepts of dialogi c relationship of Kent et al. (2002). As a resulr, OPR will be affected by web site performance according ro argumenrs ofKenr et al.  However, there were ma ny difftcul ties i n m easur ing OPR (Plowma n, 2007). Th ere are no specific methods to m easure a long-term relationship because p revio us metho ds o nl y m easured t he o ut p u r a nd o utcome of relative short-term relati onshi p. Th ey can only give info rmat1on a bout rhe effective ness of pa rticular public rela rions p rogra m (Hon & Grunig, 1 999). Wh en ir co mes ro measu ring overall effect of public relations for orga nizations, we need different rools a nd techniq ues.
Hon & G rui ng (1999) p roposed a qua nti tative scale for measuring OPR, iden ti fyin g six feat ures: (1 ) Trust: one pa r ry's con fidence to open oneself to the other; (2) Control Mutu ali ty: the degree of agreement of one party has the rightful power to affect the oth er; (3) Satisfaction: the degree of sat isfaction of on e pa r ty feels toward the ot her; (4) Commirm e n t: the degree of feel ing rhat relatio nship is wo nh spe nding e nergy to m aintain and promore; (5) Excha nge Relationshi p: one party gi ves rhe other beneftts only beca use rhe ot he r has provided in t he past or is expecred ro do so in rhe furure; (6) Comm u n aJ Rel a rionship: both par ties pr ovided ben eftts to the other beca use they a re conce rned for rhe welfare of r he other (p.3). Howeve r, most schol a rs m easured OPR only with the first fo ur i ndicators (e. g., Hua ng, 200la; Ka n g et a l., 2008; Klm, 2008; Pl owm an, 2007). Grunig & G ru n i g (2001) also clai m ed rh at "four of these are especially impor ta nt for measu r in g r he q uality of organ izationp ubl ic relationsh ips (p.24)." Kent et al. (2002) lin ked the concept of the Internet with the relationship throu gh web site design, i mplying that web si te desi gn co uld affect OPR; Hu a ng (2001 b) also fo und that OP R could red uce conflicts, gene rating cooperat ion bet we en organizati ons a nd publ ic (which can minim ize che damage of crisis). This a nide ado pted methodology bro ught up by Hua ng (2001 b), taking OPR as a mediator, which affected the needs for crisis commu nication a nd being i nflue nced by corpor ate web site performa nce (Kent et al., 2002). Accordin gly, resea rcher proposed the following hypothesis abo ut OPR: H1: OPR wi ll media re the effect of cor porate web si re perfor mance on the needs for crisis comm unication .
2.3.4. Cr is is responsi bi l ity 95 Crisis respo nsi bili ty is a n i mpor ta nt factor i n terms of discussing crisis comm u nicat ion . Based o n Coo mbs (1998a) , this a nide defined crisis respo nsibility as "the exte n t of a n organi zation's responsibility for th e crisis, w hich is an ou tcome of evaluation of public." If publ ic don't co nsider a n orga niza tio n is responsibl e for th e crisis, its corporate i mage wou ld n't be i n jured, a nd there is no needs for crisis communication ( Benoit, 1 997). On the orher ha nd, if there is a fine relat ionship betwe en orga ni zat ion a nd public, publi c wo uld assume that orga nization sho uld cake less res ponsibility for th e crisis (Coombs & Holladay, 2001 ). l n other words, crisis responsibili ty is on e of the key con cepts to associ ate relationship b uilding wi t h crisis comm u n ication. Coombs (1998a), Coombs & Holladay (1 996, 2001) took attri b utio n rh eory as a n ex pl anatio n too!, suggesti ng t h at pu bl ic eval ua re orga ni zation's crisis responsibili ty t h ro ugh its per forma nce history. Kl m (2008) also fo und that crisis responsibi li ty has a significant nega tive influence o n orga nizati o n al repu tation, and t hereby affecting parti cipa nts' pote nti a l s upporti ve behaviors (i. e., "WOM intention").
Th e crisis of toxic mi l k powder was a different situation to previous studies: I ts affected ra nge was wide-s pread, a nd m a n y enterprises and consumers were suffering. Sim ilar incidents haven't occurred before. Therefore, the influence of i n jured pa r ty (Pa rk & Len-Rlos, 2008) a nd crisis his tory (Coo mbs, 1998) will be smaller, a nd the inAuence of rela rion histo ry will be la rge r. Th e "OPR" which termed in rhis a nide acru a lly is equi vale nt to relation history a nd the di alogic relationship berween o rganizarions and pu blic above-men t ioned, bei ng b uil r thro ugh we b si te perform a nce.
Accordi ng to prev io us l i terat ure, resea rcher co nsidered rhat web si te perform a nce will affecr rhe perceived crisis respo nsi bility, and the effet of web si te pe rform a nce on the needs for crisis communication is med tated by crisis responsibili ty: H2: Crisis responsi bili ty will media re the effect of corporate we b site perfo r mance on the needs for crisis comm unicati on. 2.3.5. Sel f-efficacy Accord ing to Bandura (1982), self-efficacy is a perce pti on of interna! self-system, a n eval uation of abili ty ro complete specific performance, a nd would affect indi vidual's choice or en ergy determined to invest in. Generally, those wirh high self-efficacy will be provoked ro a greater capacity by obstacles.
Gr unig's si tua tional theory assumed t h a t public' s co mmunicario n behaviors (i. e., inform ation seeking and inform ation processing) could be u nderstood via thei r perception to environme nt, a nd could be identified as th ree indepe nd en t va riables: (1) Problem Recogniti on; (2) Constraint Recognition a nd (3) Leve! of Invol vement (Gr uni g & Hunt, 1984). Among rhese va riables, the concept of constrai nt recognition appare ntly has included the consideration of self-efficacy.
In theory of G runig, self-efficacy as one of situat ion al va riables could be constructed through cognition related to issues, making influe nce o n individual's behaviors. If attitude is formed by el a borare cogniti ve processes a nd is highly constructed, there is likelihood fo r l asting behavior ch a nge. When actions are specific and easy to impl em ent, rhe cognitive cha n ge t hrough co mmunication is likely to provoke a short-term behavior cha nge (Dozier & Ehling, 1 992). G runig & Childers (1 988) further revised si tuational theor y, assumin g that problem recognition, co nstraint recognition and leve ! ofinvolveme nt have both interna] and externa! di mensio ns (as cited in Dozier & Eh ling, 1992). In terna! constra im recognition apparentl y is adverse of efficacy expectati o n a nd externa!constraint recognitio n is adverse of outcom e ex pectatio n. That is, they m easu red self-efficacy of public by (1) efficacy expectation and (2) ourcome expectatio n (which a re critica!factors in Ba ndura's self-efficacy theory).
In che study concern ing mediation effect of perceived interactivity of So n g & Buey (2006), self-efficacy is a crucia l moderare variable which affect sequen tial behaviors. Thei r results showed rhat objective interactivi ty a nd part icipam's self-efficacy have significa nt interaction effect on perce i ved i nte r ac ti vity. Regardless th e case of low or high ob ject ive interacti vity, as self-efficacy i ncreased, percei ved interacti vi ty decreased .
The expla nation of Song & Buey is that those with high self-efficacy might experience a higher sense of i nteractivity before, thereby when they were given insufficient a mo unt of objective interactivity, their perceived interactivity decreased .
Among domestic (Tai wa n) resea rches, Lo, Li, Shi h, & Yang (2005) adopted expectancy theory as a framework to discuss che In ternet adoption and uses of public. They found tha t expectancy value is an important predictor ro both adoption a nd uses. In fact, expectancy theory a nd self-efficacy theory have much in comm on, such as t he con cep t ofinstrumentali ty a nd that expecta n cy a re si mi l ar to efficacy expectation a nd outcome expectauon.
Those studies mention ed abo ve showed r h at self-efficacy is particul ar for interpreti ng individual behaviors. Th is a nid e defined outcome expectatio n as "the belief to avoid dam age ca used by the crisis of toxic mi l k po wder through the infor matio n from the Internet", a nd efficacy expectat ion as "the abili ty to acquire th e info r mation from the Internet". Due to the fact that interactio ns between environment, individual a nd be haviors, and self-efficacy is the most importa nt regulatory mechanisms for influe ncing behaviors and also che cogni tive motivation which prom etes individual to behave, researcher considered rhat web site perform a nce (environme nt factor) would a ffecr the sel f-efficacy of consumer, thereby influencing the needs for crisis communication: H3: Self-efficacy wi ll mediare che effect of corporate web si te performance on ch e n eeds for crisis communicat ion.

Method
The experim ent method has been widely used in crisis resea rches, especially during che time when researchers want to explore the perception of public (e. g., Coom bs, 1998a;Coom bs & Holl aday, 1996, 2001Kang et al., 2008;Kim, 2008;Pa rk & Len-Rios, 2008). In arder to test research hypothesi zes, one-way (low vs. high web site perfo rmance) between s ubject design was conducted. And for investigari ng the effectiveness of experiment stimulus more clea rly, resea rch e r also iden tified related variables in previous studies.

Subjects
This research aimed at recruiting young participants th at have high homogeneity in age and educacional background in order to meet research needs and to increase interna! validity. Researcher recruited voluntary online participants from Bulletin Board System (BBS) sites (Ptt and Ptt2, having enrollm ent of about 100 milJion people, which are the biggest two si tes in Taiwan). The average age ofBBS users were 24 years-old, who have considerable knowledge abour rhe Internet. They are rhe biggest online group in Taiwan. Oue ro rhe fact that the crisis of toxic milk powder is a n issue relevant to general p ublic, wo uld have less i mpact with the characteristics of BBS users.
Participants for rhree testing in this st udy were all voluntary response sample. There were 43 parricipants i n flrsr pretest (between subject desigo). They were mainly tested the man i pulation of stimulus and internal-consistency rel i abi l ity of questionnaire items, including OPR, crisis responsibility, needs for crisis communication and individual relevance ,etc.. There were 25 participants in second pretest (within s ubject desi go), being tested manipulation ch eck a nd revised self-efficacy items.
Finally, the participants (N=402) in formal experi ment, with no one ever participated in pretest, were random l y assigned and ex posed ro either low (n=209) or high (n=193) web site perform ance cond ition. The participants ranged in age from 1 5 to 53 yea rs (M=23.97, 50=3.72).

Variables
The independent variabl e research er manip ul ared is web site perform ance, including rwo conditions: high vs. l ow. The total score of high web si te performance on six di mensions of web si te desi gn were higher rhan rhe low. For meeting acrual cond i tion and making experiment more realistic, researcher replica ted pre-existing corporate si tes as stimulus.
However, according ro relationship managemenr approach, existed repu tatio n of an organization may affect stock value, employee's morale or related eval uarion, etc. Organi zation reputation is formed by public a nd is affecred by many facrors, includi ng service quality, eco nomic performance an d business srrategies. Previous researches indicated rhat organization reputation wo uld affect pa rricipants' attitude a nd eva luation of crisis responsi bi li ty for organizarions (Kim, 2008;Lyon & Cameron, 2004), a nd OPR wou l d affect organization reputation (Coombs & Holladay, 1996, 2002. As a result, researcher took organ ization reputation into control.

Organizat ion reputatio n
Most studies measured organization reputation using 5-item scale from Coombs & Holladay (2002) or 10-item scale from Coombs & Holladay (1996). However, rhe definition ofFediuk, Buddenhage, Masen, & Botero (2008) were more integrat ive. They defined organ ization reputation as "a generalized, overal l eval uation about an organization, the notion that these overall evaluations ca n be captu red by collapsing image components across tapi es and stakeholders, a n d the addirion of time ro the definirían (p.l81)." Organization reputation provides the information about furure performance and is develo ped over time. Accordingly, researcher developed 4item scale, chis scale ranged from 1 (strongly disagree) to 5 (strongly agree), che same to all other items, except for subject variables. Befare visiting web si te assigned, participants need to respond to chis 4-item scale first.
Crisis responsibili ty was measured using 2-item scale from previous stud ies (Kim, 2008;Park & Len-Ríos, 2008), asking if an organization is responsible and the extent of an organizarían need to be blamed. And Self-efficacy was measu red using 6-item scale regarding to the crisis of toxic milk powder. In this scale, 3-item was about outcome expectation, and other 3-item was about efficacy expectation.

Needs for crisis communication
The definition of needs for crisis communication ís "the characteristic ofinformation which public pay attentíon ro during the crisis." According to previous studies (Sturges, 1994;Taylor & Kent, 2007) and observarían in thís case, researcher developed a 6-item scale: (1) the amount of crisis communication messages; (2) consistency, (3) immediacy, (4) sígnificance, (5) usefulness and (6) credibility of crisis communication messages during the crisis.

Subject variables
Internet usage behavior, trust on web information and personal relevance were the critica!variables in the Internet related researches (Aspden & Katz, 2000;Green, 2007), this anide also included related variables in questíon naire.
Internet usage behavior measured using 3 indicators. Trust on web informarían means "the extent of trust on informarían from the Internet ", including 9 indicators. Personal relevance were measured with 6-item about recognitíon from Kang et al. (2008), and researcher added 4 items on behavior and outcome. For the reason that this article adopted preexisting corporate si tes as stimulus, there could be an effect of prior knowledge of participants for stimulus. In ard er ro partía! out rhis influence, researcher meas ured browsing frequency with one ítem: "Before responding to this questionnaíre, what ís your daily bro wsing frequency of this corporate web site?" Answer ranged from 1 (never) to 6 (always). The following demographic inform a rían from participanrs was collected: sex, age, education a nd incom e, beca use they were also important predictors (Aspden & Katz, 2000;Lo et al., 2005).

Experiment Procedure
According to the result of second pretest, researcher adopted the corporate web site ofYakult (high) a nd Costeo (low) in Taiwan as stimulus. And, in arder to rule out the influence that Yakult's site communicated for the crisis but Costco's didn't, researcher removed rhe crisis comm unication message in Yakulr's site.
Alrhough the nationality a nd business category of corporat ion would affect crisis responsibility, researcher still didn't include into control. There were two reasons: (1) incompatible with research questions and (2) the difficulty of implementation.
The formal experiment was conducted from Apr. 2 ro 13 2009 in Taiwan. Each participant was assigned to either Yakulr's (h igh performance) or Costco's (low performance) si te, and compl eted the questionnaire after browsing the site (browsing tim e was about 1 min average).

Reliability Analysis and Manipulation Checks
The reliabiliry analysis of ove rall items produced an acceptable interna!consistency (the value of Cro nbach's _al! above 0.70), see rabie l.
To assess the effectiveness of experimental manipulations, a series of one-way analysis of var iance (ANOVAs) were conducted. ANOVA sho wed that there were significant differenc es for web si te performance between two groups, F(l,400) = 123.97, p<O.OOl., and there were no significant differences for o rganization reputation, F(l ,400) = 0.31, n. s.. As expected, the web si re performance score ofYakult's si te (M=3.67, SD=0.44) was significan tly higher rhan Costco's (M=3.09, SD=0.58), and there were no differences in rheir organizarían reputation. Overall, the analysis of the manipul a rían check indicated the fact rhat the intended manipulations were successful (see ra bie 2).

Test of Subject Variables
For ensuring the characteristics of participants from two groups were no differences, Chi-square test of independence and ANO VA were co nducted. Results showed that rhere were n o signifi ca nt differences for su bject variables between two groups, except for browsing frequency, F(1,400) = 40.84, p<O.OOl. The freque ncy of Costco's (M=2.11, SD=0.06) were higher than Yakult's (M=1.54, SD=0.06). While the distribution ofbrowsing frequency of two groups was not the sa me, ir should be due ro the selection bias.

Test ofHs
To address Hs (which presented the mediarían relationship berween variables), a series of multiple regression a nalysis (steps ofBaron and Kenny, 1986; as cited in Kenny, 2008) were conducted usi ng web site perform ance as independent variable, OPR, self-efficacy and crisis responsibility as mediators and needs for crisis commun ication as dependent variable. In order ro use bootstrapp ing which is a better approach for testing indirect effect, researcher adopted the SPSS macro ofPreacher and Hayes (2008).
First, results showed that the model already has a signiflca nt total effect to be mediated in path C, B = 0.130, SEB = 0.050, _ = 0.128, p<0.05.
Step 1 was supported.
Next Step 3 was parrly supported. Finally, for path C', there was no signiflcant direct effect of web si te performance on needs for cris is communication when mediators was controlled in regression model, B = 0.064, SEB = 0.045, _ = 0.063, n. s.
Step 4 was supported. Res ults of steps were summarized in Table 3.
In rhe multiple mediarían model, a signiflcant regression equation was found with adjusted R_ of0.237, F(4,397) = 32.133, p<O.OOl. When taking mediators into model, the change with adjusted R_ (0.223) was significant (p<O.OO1). Furthermore, paths of mediation model were illustrated in Figure l.     As ex pected, results supported hypoth esize: influential mechanism ofweb site performance on needs for crisis communication is not simply direct effect but a mediation effect. Results indicated tha t Hl (O PR) and H3 (self-efficacy) were supported, while H2 (crisis responsibility) wasn't confirmed.

Discussion
Over the past two decades, Huang (2001 b) suggested, the effect of public relations for organ izations has always been one of the hot tapies. Huang's study revealed the i mportance of relarionsh ip manageme nt. Previ ous srudies concerning Interne t PR were based on an assumption that seeing the Internet eq uival ent to one of traditional communication tools, however, merely examining its a uthent icity.
This article examined this assumption, supporting the argum ent of previous studies (Kent & Taylor, 1998;Taylor et al., 2001;Kent et al., 2002;Sun, 2004). The difference between t raditional and n ew media on plarform features or whose role as a comm u nication too! is m ainstrea m, relationsh ip has always been the core value of com munication and the important topic in public relat ions.
Research resul ts indicated that Internet PR is helpful for organizations. When the web site perform ance of a co rpor ation is good, its crisis responsi bi li ty wo uld also be decreased. Th is article also provides the a nswer ro questions of Hill & White (2000): What is the effect of the Internet PR? How to assess its effect? For organizations, the effect of Internet PR is building a nd maintaining good relationship, the same to rraditional PR. Through the way of direct link with public, web site perfo r mance has a positive effect on OPR, and i ts measur es could be the assessment indicator.
Note th at corporat i ons wh ich have better web si te performa nce shouldn't ignore the import ance of crisis comm unication. Previous researches indicated tha t good OPR could decrease crisis responsibility for organ izations and promote publ ic's potencial suppo rtive behaviors (Kang et al., 2008;Ulmer, 2001), seemed ro i mply rhat corpor ations needn't ro implement crisis communicat ion and will have a good res ult. In rhis study, good web si te perform ance led to better par ticipa nts' self-efficacy, thereby di rectly and indirectly increased participants' needs for crisis communication. Although th ere is less crisis responsibility, i t doesn't mean r h a r needs for crisis comm unication will also be less. If corporations couldn't sa tisfY the needs of pu blic, rhere could be a negarive effect on corporate image (Benoit, 1 998;Fearn-Banks, 2002).
Since researche r ado pted p re-existing sites as stimulus, there were sorne limita tions to th is study. The diffe rence in business category berween two stimulus sites co uld affect t he outcome, a nd rhat the dependent var iable couldn't be linked ro crisis response straregy which is critica! in rhetorical approach (e. g., theo r y of image restoration; Benoit, 1998;Coombs, 1998a).
However, in order to decrease public's uncertainty to stimulus and ro more meet the actual conditio n , researcher still s uggesrs rhat future study should use pre-existing sites as stimulus. In order to partial out the effect of prior knowl edge, f uture researches could tal<e it into co ntrol or discuss the relationshi p among prior kno wledge, OPR, self-efficacy a nd crisis responsi bi li ty.