Downsizing and Alcohol Use: A Cross-lagged Longitudinal Analysis

Downsizing has become an established part of the management repertoire. One area of concern is the possible linkage between downsizing events and alcohol use and abuse; however, work in this area offers inconclusive evidence of a relationship between layoffs and alcohol use. Using a three-wave panel sample (N=455) from a large U.S. industrial firm that has undergone numerous downsizing events in the last decade, four alternative causal hypotheses using fully cross-lagged three-wave mediational latent factor models were compared via Structural Equation Modeling. Separate models were analyzed for layoff experience and job security perceptions. We found large autoregressive effects for problem alcohol use in the stability models. One reason for weak support for spillover models may be that problem alcohol use is very stable over time. All alternative causal models fit the data well; however, the only model comparisons showing significant improvement over the stability models were those including reverse paths. Reverse causal models may be plausible when examining the link between alcohol problems and the workplace. Additional theoretical and practical implications are discussed. DOWNSIZING AND ALCOHOL USE 3 Downsizing has become an established part of the management repertoire in response to increased global competitive pressures and economic challenges. For example, 90% of medium and large firms have downsized in last several years (Cameron, 2001) and repeated downsizing is not uncommon (Bureau of Labor Statistics, personal communication, June 13, 2003). As this practice becomes entrenched as a standard corporate strategy, practitioners and investigators have become concerned about the impact on employees: what are the economic, mental and physical costs not only for those who lose their jobs, but those who remain? Numerous articles confirm that downsizing not only takes a toll on those laid-off, but on survivors as well (Brockner, et al., 1988; Ferrie, Shipley, Marmot, Stansfeld, & Smith, 1998; Grunberg, AndersonConnolly, & Greenberg, 2000; Iverson & Sabroe, 1988). Those who survive repeated rounds of layoffs may pay a particularly high price (Armstrong-Stassen, 1997; Moore, Grunberg, Greenberg, 2004) as organizations continually adjust to the demands of the environment. One area of concern to employers, health care providers, and investigators is the possible linkage between these downsizing events and alcohol use and abuse. Data suggests that alcohol abuse continues to be a significant and, perhaps, growing problem in the U.S. and the workplace (Bernstein & Mahoney, 1989; National Institute on Drug Abuse, 1996; Schmid, 2003). Alcohol consumption is determined by a variety of factors, but specific influences of the work environment on abuse and problems has been of interest in the last decade (Cooper, Russell, & Frone, 1990; Harris & Fennel, 1988; Harris & Heft, 1992; Martin, 1990; Ragland & Ames, 1996) The question addressed here is if the experience of downsizing or related perceptions of job insecurity might heighten the use and, possibly, abuse of alcohol as a coping strategy. Specifically, the purpose of this study is to examine the layoff/alcohol relationship in the context of a three-wave panel study utilizing Structural Equation Modeling (SEM) to assess the temporal DOWNSIZING AND ALCOHOL USE 4 relationships among key variables of layoff contact, job security, escape motivations for drinking, and problem alcohol consumption. Downsizing and alcohol use –weak relationships? Stress-relief drinking has been identified as one of several prominent factors in the etiology of alcohol abuse and dependency (Powers & Kutash, 1985). In the corporate context, it is thought that workers drink to reduce negative affect associated with unpleasant or stressful working conditions (Cooper, Russell, & Frone, 1990; Ragland & Ames, 1996). It is plausible that downsizing and layoff events would be perceived as stressful or at least somewhat unpleasant experiences for employees and that alcohol use would increase for those impacted by these events. However, recent work in this area suggests weak or inconclusive evidence of a relationship between layoffs, or most workplace stressors for that matter, and alcohol use. For example, Steffy & Laker (1991) note that workplace stress did not explain drinking behavior; however, excessive workload did contribute to use of alcohol as a coping mechanism. They generally describe their results as “weak” evidence of a relationship between workplace stress and alcohol problems. Using partial correlations and controlling for Time 1 levels of alcohol-related dependent variables, Moore, Grunberg, & Greenberg (2003) found that relatively few work attitudes predicted the outcomes of escape drinking reasons, alcohol consumption, and alcohol problems. Johnson & Pandina (1993) in their examination of alcohol use, stress, and coping strategies concluded that there were no significant differences in patterns of relationships (e.g., coping strategies or alcohol use) between those for whom stress preceded their alcohol problems as compared to those whose alcohol problems preceded reported stress. They suggest that a direct effect between stress and alcohol-related problems is questionable. DOWNSIZING AND ALCOHOL USE 5 That so few studies substantiate a stress relief or stress reduction hypothesis is puzzling. If consumption of alcohol is a socially acceptable means of reducing stress (Seeman & Anderson, 1983; Cooper, Russell, & George, 1988; Trice & Sonnensthul, 1988; Cooper, Russell et al. 1990; Trice 1992), why are we not seeing more evidence of increased alcohol use by employees who have survived not only one, but perhaps several downsizing events? The use of more complex mediated or moderated models is moving inquiry into more productive arenas (Frone, 1999; Frone 2003; Grunberg, Moore, Anderson-Connolly, & Greenberg, 1999; Grunberg, Moore, & Greenberg, 1998; Moore, Sikora, Grunberg, Greenberg, in press) as is the recognition that relationships may not be purely linear (Moore, Grunberg, & Greenberg, 2000). We suggest that another reason for weak effects may be prevailing assumptions regarding causality or temporal direction of relationships. Most studies examining the linkage between the workplace and alcohol use assume temporal precedence of the workplace. That is, stressors in the workplace are implicitly or explicitly hypothesized to cause alcohol problems, not vice versa. Generally, this assumption remains untested because most analyses have relied on regressionbased cross-sectional or two-wave incomplete panel analyses, thereby limiting the ability to explore causal relationships. Zapf, Dormann, & Frese (1996) and others (de Jonge, et al, 2001; de Lange, Taros. Kompier, Houtman, & Bongers, 2003; Farrell, 1994; Finkel, 1995) call for more deliberate consideration of causality and review the relative strengths and weaknesses of various forms of longitudinal analyses. Virtually all end with the recommendation of fully cross-lagged panel models analyzed via SEM. These models provide an assessment of multiple dependent variables and direct/indirect effects, examine fit for the total system of effects as opposed to testing independent regression equations, explicitly model measurement error and omitted or DOWNSIZING AND ALCOHOL USE 6 common causes, allow analysis of multiple waves of data and more than two variables, and, importantly, allow the investigator to explore plausible alternative explanations, such as reversed or reciprocal effects. Following recommendations of de Lange, et al (2003) and others (de Jonge, et al, 2001; Zapf, 1996), we systematically test competing structural models to explore plausible causal linkages between layoff experiences/perceived job insecurity and alcohol problems. Specifically, we will compare a “standard” model (paths leading from organizational characteristics to employee attitudes or behaviors) with reversed (path leading from employee to organization) and reciprocal (paths reflecting mutual influence) alternatives. While multi-wave longitudinal panel designs cannot conclusively demonstrate causality (Burkholder & Harlow, 2003; Farrell, 1994), this approach to permits us to explore and compare plausible causal hypotheses. The following section will first overview how we conceptualize our independent construct of downsizing experience and describe escape motivations as a mediating factor. We will then outline the rationale for hypotheses regarding standard, reverse, and reciprocal relationships between downsizing experiences and problem alcohol use. Downsizing and Alcohol Use – Plausible Alternative Hypotheses Independent variables. We conceptualize survivors’ downsizing experience in two ways. First, a more objective approach is to focus on what we refer to as layoff contact: has the employee him/herself had direct experience with a layoff, indirect experience via coworkers or friends being laid off, or minimal experience by virtue of simply being in company undergoing layoffs. Second, we consider a more subjective response by looking at perceptions of job security. Layoff contact and job security are likely closely intertwined. Indeed, investigators DOWNSIZING AND ALCOHOL USE 7 have found that employees who have experienced layoffs anticipate job loss in the future (Ellwood, et al, 2001; Kalimo, Taris, & Schaufeli, 2003). Moore, Grunberg, & Greenberg (2004) note that all employees in their study appeared to growing less secure as they experienced repeated waves of downsizing. Exploring models using layoff contact or perceived job security as independent constructs will offer insight into how specific objective experience (layoff contact) and more general objective perception (job security) relate to problem alcohol use. Mediating variable. Following recommendations to examine more complex models of the workplace-alcohol connection (Cooper, Russell, & Frone, 1990; Grunberg, Moore, Anderson-Connolly, & Greenberg, 1999; Harris & Fennell, 1988; Harris & Heft, 1992) as well as calls for longitudinal modeling that explicitly accounts for third variables (Zapf, et al, 1996), we incorporate a motivational construct, escape reasons for drinking, as an intervening or mediating variable between downsizing experiences and problem alcohol use. There is an established body of research that suggests that those individuals who drink to cope with or escape stressful situations are more likely to be problem drinkers (Abbey, Smith, & Scott, 1993; Cooper, Russell, & George, 1988; Farber, Khaven, & Douglass, 1980; Grunberg, Moore, Anderson-Connolly, & Greenberg, 1999; Martin, Blum, & Roman, 1992). Most work in this vein has conceptualized escape reasons for drinking as a moderator or an individual difference “trait” that is stable across time. However, based on a motivational model of alcohol use (Cooper, Frone, Russell, & Mudar, 1995), avoidant coping motives could be viewed as more proximal to alcohol use and abuse. In this framework, escape reasons are conceptualized as malleable or situation dependent rather than fixed: if the environment does not stimulate a stress response, the need to engage in coping behaviors should not arise. DOWNSIZING AND ALCOHOL USE 8 In a preliminary analysis of two waves of data controlling for previous levels of drinking, we found that escape reasons for drinking predicted alcohol consumption. In addition, Grunberg, Moore, and Anderson-Connelly (1999) found that workplace experiences and attitudes, e.g., perceptions of stress or overload, were associated with propensity of employees to endorse escapist reasons. Moore, Sikora, Grunberg, and Greenberg (in press) also found evidence for a mediational path for some groups of workers. These data were analyzed using an incomplete lagged panel or cross-sectional approach, however, so questions of causality or temporal order were left open for subsequent exploration. The question addressed here, then, is not if escape motivations set the stage for differential interpretation of work experiences and subsequent alcohol use (a moderation question), but if work experiences play a causal role or are related to changes in escape reasons for drinking which, in turn, are linked with different alcohol use outcomes (a mediation question). This question can be formalized as: H1: Layoff contact and perceived job insecurity will be significantly associated with alcohol problems (zero order relationships help establish mediation, per Kenny, 2001). H2: Escape reasons for drinking will mediate relationships between layoff contact and job security and problem alcohol use. Specifically, we will see evidence of mediation within waves as well as between waves (suggesting causal as well as mediational relationship). A standard causal hypothesis. As noted earlier, the prevailing assumption in occupational psychology (e.g., Hackman & Oldham, 1980; Karasek & Theorell, 1990) reflect what de Jonge , et al (2001) would refer to as a standard model of causality where events in or characteristics of the workplace are posited to subsequently cause worker distress and poor outcomes. Spillover and stress relief hypotheses currently dominate the literature regarding work stressors and DOWNSIZING AND ALCOHOL USE 9 alcohol problems and are exemplars of this standard framework. Specifically, the spillover hypothesis suggests that negative experiences in the workplace generalize or “spillover” into non-work domains and affect employee well-being and behavior (Greenberg & Grunberg, 1995; Jahoda, 1982; Kohn and Schooler, 1983; Karasek & Theorell, 1990.) In the alcohol domain, negative experiences are expected to spillover into problem alcohol use as employees turn to alcohol as a self-medication strategy to deal with stress. (Cooper, et al, 1990; Greenberg & Grunberg, 1995; Martin, Blum, & Roman, 1992; Trice, 1992; Trice & Sonnensthul, 1988). This is consistent with stress reduction hypotheses that suggest that stress in the workplace causes workers to look alcohol as a coping mechanism or means to reduce discomfort or negative affect (Cooper, Russell, & Frone, 1990; Powers & Kutash, 1985 Ragland & Ames, 1996) While there is some evidence for resilience or adaptation as individuals learn to cope with repeated stressors (Grunberg, Moore, & Greenberg, 1998), the stress vulnerability hypothesis has received support in downsizing contexts. That is, rather than distress plateauing or perhaps declining as workers adjust to workplace changes, evidence suggests that workers become increasingly vulnerable to negative emotional and health consequences over time (Grunberg, et al., 1998, 2001; Moore, et al, 2004). This pattern has also been noted with job security where chronic levels of job insecurity have been found to have a negative effect on employee wellbeing (Ferrie, Shipley, Stansfeld, & Marmot, 2002; Heaney, Israel, & House, 1994). The combination of standard spillover or stress reduction hypotheses with a stress vulnerability orientation suggests that: H3 (Standard hypothesis): Layoff contact and job insecurity will increase escape motivations for alcohol use and increased alcohol problems over time. DOWNSIZING AND ALCOHOL USE 10 A reverse causal hypothesis. While the standard models dominate current work in this area, it is plausible to consider a reverse hypothesis in which worker behaviors or attitudes cause changes in their work environments or, at least, their perceptions of their work environments. A review by Zapf, et al (1996) of 45 longitudinal studies found 6 of 16 “high quality” investigations with evidence of reverse causation. They conclude that while not always strongly supported, reverse effects should be considered a plausible alternative model as few content areas have explicitly disconfirmed this alternative. The so-called drift hypothesis (Frese, 1982; Kohn & Schooler, 1983; Lorence & Mortimer, 1985; Williams & Podsakoff, 1989) suggests that more motivated and emotionally or physically well individuals drift toward better jobs, are more apt to perform well on the job, or have fewer absences or job incidents, thereby reaping rewards associated with consistent or high performance (Daniels & Guppy, 1997). Conversely, those with poor attitudes, or perhaps poor health, find themselves in less appealing situations or viewed as “poor performers” by their peers and/or supervisors. In a downsizing situation, these individuals may be explicitly targeted for layoffs or, perhaps, less apt to receive support or advocacy when human resource decisions are being made. Similarly, an individual who is in poor physical health or has certain psychological predispositions (depression, negative affect) may perceive the environment to be more threatening or less secure as compared to peers who are relatively healthy or positively oriented. Social cognition literature regarding schema’s (Markus, 1977; Fiske and Taylor, 1992) suggests that believing is seeing: individuals with a more negative orientation or belief system regarding their workplace may selectively attend to or recall elements that conform to their beliefs about how bad things are or how tenuous their position is at the firm. Even though the environment may be DOWNSIZING AND ALCOHOL USE 11 neutral or even positive, the employee’s perception is negative. Self-fulfilling prophecy (Merton, 1972; Madon, et al, 2001) suggests that these employees may ultimately experience the negative outcomes they envision. Based on the above, we believe that a reverse hypothesis is plausible. Specifically, H4 (Reverse hypothesis): Problem alcohol use will increase the likelihood of escape motivations for drinking, decrease perceptions of job security, and will also increase likelihood of layoff contact over time. A reciprocal causal hypothesis. While not often tested, the idea that characteristics of the job and employee response are mutually reinforcing is not an uncommon theoretical position (Edwards, 1998; Williams & Podsakoff, 1989, Zapf, et al, 1996). In these theories, the response or behavior of the employee is anticipated to influence work conditions (or perceptions of them) and stressors associated with these changing conditions exacerbate these perceptions, behaviors, or responses. For example, an individual who struggles with asthma may miss work more often as a result of respiratory distress. This employee, in turn, might be viewed as unreliable by peers and excluded from meetings or important projects. The stress of exclusion heightens the asthma response, which in turn, increases lost work time. A vicious cycle or feedback loop ensues where the employee unwittingly or unwillingly perpetuates poor personal and organizational outcomes. While very little empirical work has been conducted in this vein, we include this reciprocal hypothesis in order to complete our systematic review of plausible alternatives (Zapf, et al, 1996): DOWNSIZING AND ALCOHOL USE 12 H5 (Reciprocal hypothesis): Layoff contact, job security, escape motivations, and problem alcohol use will be mutually reinforcing such that downsizing experiences will both cause and be caused by alcohol problems and escape reasons for drinking. The following figure summarizes the conceptual framework and hypotheses. Insert Figure 1 about here Methods Background The Research Site. The research was carried out in a large, manufacturing division of a multinational company on the west coast of the United States that produces sophisticated high technology products. The company’s employment has fluctuated between 150,000 to 200,000 employees over the past decade, with the division’s employment ranging from around 60,000 to 120,000. Employees in the division experienced two major episodes of layoffs and rehiring in the period 1995 to 2000, with contractions of 25 to 30 percent in the workforce soon followed by similar sized expansions. The most recent episode of layoffs occurred in 2002 and 2003 and affected some 35% of the workforce (or close to 35,000 persons). Both white and blue-collar workers across a wide spectrum of occupational skills and organizational positions -including managers, professionals, design engineers, lower level clerical workers, and semi-skilled machine operators -are represented in the organization and in our sample. Sample. In the winter of 1997 a random sample of 3,700 employees was obtained to participate in a longitudinal study that examined the effects of workplace change on job attitudes, work experiences, mental health and physical wellbeing. Workers who responded at Time 1 (n = 2279; 62% response rate) and who were still employed by the company two and one-half years later (n = 1960), were contacted a second time in the fall of 1999 and again asked to participate. DOWNSIZING AND ALCOHOL USE 13 Similarly in 2003, we contacted participants who participated at both Times 1 and 2 (n = 1244, 63% response rate) and who were still employed with the company (n = 1046) and asked for their participation a third time. A total of 776 employees responded at Time 3 (74% response rate). At each time period, respondents were contacted with a letter sent to their home and asked to participate in the study by completing an enclosed survey. They were assured that their responses would be confidential and that they would be paid ($20.00 Waves 1 and 2; $35.00 Wave 3) for their participation. Before each survey administration, we conducted a number of individualized interviews and focus groups in order to familiarize ourselves with recent changes in workplace practices. A postcard reminder was sent to non-respondents, followed by an additional mailed questionnaire and second postcard reminder in an effort to increase the response rate. After list-wise deletion of missing data, the final sample for this three-wave longitudinal analysis is 455. Sample loss is primarily due to abstainers: individuals who report not drinking at all during each survey time period are skipped out of questions regarding motivations for and consequences of alcohol use. The final sample, then, includes only those individuals who report consuming alcohol for each of the three waves of data collection. Measures Layoff contact (Layoff1, Layoff2, Layoff3). This single item latent measures three levels of layoff experience: none, indirect, and direct. Grunberg, et al (2001) found that any type of contact with layoffs was associated with negative outcomes. However, they also found that those with direct contact, i.e., being laid-off and rehired or receiving a warn notice that they were a target for potential layoffs, had even more pronounced negative effects than those who had only indirect exposure, i.e., friends or co-workers were laid-off or warned but they themselves were DOWNSIZING AND ALCOHOL USE 14 not directly impacted. This scale accounts for a broader range of respondent experience than does a simpler “were you laid-off, yes/no” item. Job Security (JobSec1, JobSec2, JobSec2). This three item latent measures overall sense of job security. The three items encompass past, present, and future security expectations. For example, respondents were asked how worried they were about job security over the previous two year period (the same time period in which layoff contact might have occurred), how worried they are today, and then, how confident they are that they can remain with their employer for as long as they want. Each item utilizes a four point scale. The first two items use response options of extremely worried, somewhat worried, not very worried, and not at all worried; the last item uses options of very confident, confident, not confident, and not confident at all. Escape Reasons for Drinking (Escape1, Escape2, Escape3). This latent variable measures the degree of importance people give to four “escapist” reasons for drinking. These include drinking to “relax” and drinking “to forget about” one’s job. Response options were not important, somewhat important and very important (Fennell, Rodin & Kantor, 1981). Alcohol Problems (Alcprob1, Alcprob2, Alcprob3) is a latent variable comprised of three indices. CAGE is a standard measure of alcohol dependence which asks respondents to answer yes or no to whether in the last 5 years (a) they have felt they “ought to cut down” on their drinking; (b) people have “annoyed” them by criticizing their drinking; (c) they have “felt bad or guilty” about their drinking; and (d) they have had a “drink first thing in the morning” to steady their nerves or get rid of a hangover (Ewing, 1984). Negative Consequences from Drinking is a modified, five-item version of a scale developed to measure consequences of drinking alcohol (Jessor, Donovan & Costa, 1991; Cahalan, 1970). The items asked respondents how many times DOWNSIZING AND ALCOHOL USE 15 (ranging from never to four or more times) in the past 12 months the following things happened because of their drinking: they were criticized by friends; they missed work or had to call in sick; they had difficulties with spouse or partner; a family member expressed concern; they had driven when they had had a good bit to drink. A square root transformation was employed to correct for positive skew (Judd & McClelland, 1989; Kline, 1998). Alcohol consumption was measured by asking participants to indicate the number of times in the past six months, ranging from never to every day, they had drunk four different amounts of alcohol: (a) one to two drinks, (b) three or four drinks, (c) five to seven drinks, (d) 8 or more drinks. These quantities were multiplied by the number of occasions to arrive at a total number of drinks consumed in the past 6 months. Variance of this indicator was very large relative to other items in the analysis so to prevent convergence problems, we multiplied scores for this item by a constant (.1) (Kline, 1998) and also utilized a square root transformation to adjust for positive skew. Analysis Approach As recommended by Zapf, et al (1996) and de Jonge, et al (2001) we conducted systematic evaluation of models in which a baseline model including only stability paths was compared to more complex models incorporating cross-lagged paths as outlined in the hypotheses. When discussing the results, the models will be referred to as: L1/J1 = stability or baseline model – no cross lagged paths, only stability paths; L2/J2 = standard or spillover model – “down” cross lagged paths in addition to stability paths; L3/J3 = reverse or drift/schema model – “up” cross lagged paths in addition to stability paths; L4/J4 = reciprocal or cybernetic model – both up and down paths in addition to stability paths; with L referring to the Layoff Contact models and J referring to the Job Security models. DOWNSIZING AND ALCOHOL USE 16 Based on the recommendations of Vandenberg and Lance (2000), we used TLI, CFI, and RMSEA in addition to the CMIN/df to evaluate fit of individual models. Values at or under 3.0 for CMIN/df, .90 or greater for TLI and CFI, and .05 or less for RMSEA are usually interpreted as “good” fit. The chi square difference test was used to assess change in fit upon release of constraints (Kline, 1998). The overarching goal in adding paths is to balance parsimony with fit (Kline, 1998). If paths are added and fit remains statistically equivalent to the more parsimonious baseline model, we choose the more parsimonious baseline model as the addition of cross lagged paths does not improve the model over and above the stability paths. Initial comparisons were conducted by adding cross-lagged paths as a unit, i.e., multiple degree of freedom tests. Based on these results, additional single degree of freedom tests were deployed to identify the “best” model based balance of fit and path significance. In the results, this model is referred to as L5/J5. A covariance matrix was submitted to AMOS 5.0 using maximum likelihood with listwise deletion of missing cases. No convergence or identification problems were evident. Per standard practice, one indicator per latent was fixed to 1 to scale the latent construct and, because it is assumed that a single item latent is measured without error (Garson, 2003; Kline, 1998), error variance is fixed to zero for latents measured by only one indicator (LAYOFF1, LAYOFF2, LAYOFF3). A list and description of variables (indicators and latents) used in the analysis is found in Table 1. Following the two-step approach advocated by Anderson and Gerbing (1988), we first evaluated measurement models for convergent and divergent validity before testing structural models. Recall that two separate models were evaluated: one using layoff contact as the independent variable and one using perceived job security. Apart from DOWNSIZING AND ALCOHOL USE 17 inclusion of escape reasons for drinking no other controls or third variables were explicitly included in the models. Following recommended practice in longitudinal SEM, all models include measurement error covariances across time based on the assumption that errors of repeated measures covary (Burkhoder & Harlow, 2003; Kline, 1998). Also, all covariances for exogenous and disturbances were modeled regardless of statistical significance to account for common causes that are not included in the model as well as synchronous relationships within waves. Results Measurement models Established practice in SEM requires that we first confirm convergent/divergent validity for measurement models before examining structural models. Standardized factor loadings for all multi-item latent factors exceed .50 suggesting reasonable convergent validity (Kline, 1998). Figure 2 summarizes indicators and loadings for the job security model; we do not show a figure for the layoff contact model as loadings for escape reasons and alcohol problems are the same as the job security model, and, because of the constraints placed on the single indicator layoff contact latent, loadings are 1.0 (perfect measurement assumed). FIGURE 2 about here Divergent validity is usually confirmed when inter-latent zero-order coefficients are less than .85 (Kline, 1998). Table 2 summarizes these zero-order values. TABLE 2 about here The inter-latent correlations among the three wave alcohol problem factors approach the threshold with a .84 (p < .001) between Wave 1 and Wave 2 and.83 (p < .001) between Wave 2 and Wave 3 alcohol latents. The correlation between Wave 1 and Wave 3 alcohol problem DOWNSIZING AND ALCOHOL USE 18 factors is not as high but is still strong at .74, p < .001. In general, with the exception of the three layoff contact factors, between wave correlations for the same construct tend to be quite high. We will further address the implications of these strong relationships in the next section when we discuss stability path coefficients. Finally, in regards to our hypotheses about a mediating role of escape reasons for drinking (H1 and H2), we conclude that there is not a mediating role as there is essentially no relationship to mediate (Kenny, 1998, 2001) within or between waves. That is, within and between all waves we see very modest or nonsignificant correlations between the two independent constructs, layoff contact and job security, and the outcome variable of problem alcohol use. As we will discuss later, escape reasons clearly are associated with problem drinking, and, to a limited degree, there appears to be an association between escape reasons for drinking and perceptions of job security. However, lack of strong zero order relationships between layoff contact, job security, and alcohol problems prevent us from confirming a mediational role for escape reasons for drinking. Model comparisons Stability models. Before addressing fit of the hypothesized alternative models, it is helpful to look first at the baseline or stability models because discussion of any cross-lagged effects presumes “over and above” variance explained by autoregressive effects (e.g., stability of the construct across time). That is, without these controls, we are likely to see apparently strong, but spurious, cross-lagged relationships between variables of interest (MacCallum & Austin, 2000). Perhaps the most important findings of this analysis are the very strong stability coefficients for problem alcohol use, and, to a lesser degree, escape reasons for drinking and job DOWNSIZING AND ALCOHOL USE 19 security perceptions. Tables 3 and 4 provide a summary of structural path coefficients for all models evaluated for this analysis (Layoff Contact and Job Security models respectively). Tables 3 and 4 about here Because stability coefficients for problem alcohol use escape reasons for drinking are the same between the layoff contact and job security models, we refer to Table 3 for a discussion of these results. Looking at the top section of Table 3 labeled L1-Stability Model, we see that standardized path coefficients for problem alcohol use are .84 (p < .001) and .83 (p < .001) between Waves 1 and 2 and Waves 2 and 3 respectively. In addition, the coefficient between Wave 1 and Wave 3 problem alcohol use is very high (not shown on model, but exceeds .75). These coefficients suggest that problem alcohol use is very stable over time leaving relatively little variance to be accounted for by other factors, work or otherwise. Similarly, path coefficients for escape reasons for drinking (.73 Wave 1!2 and .77 Wave 2!3, p < .001) suggest that employees who were deferring to alcohol as a coping mechanism in Wave 1 were doing so at Waves 2 and 3, again leaving little room for other factors to influence these motivations. The actual experience of layoff contact was not stable across waves. Standardized path coefficients were quite low (.20, p < .001, W1 ! W2; .16, p < .001, W2 ! W3) indicating that each downsizing event was different for most employees; e.g., Wave 1 might have brought a warn notice, but Wave 2 may have found the same employee somewhat removed from the event with “no” contact. However, when reviewing stability coefficients for the job security models (Table 4, top section), we note that while layoff experiences differed over waves, perceptions of job security appear to be relatively stable with path coefficients between both waves at .67 (p < .001). This suggests that those employees who felt relatively secure or insecure at Wave based DOWNSIZING AND ALCOHOL USE 20 on fit and path significance 1 tended to feel the same at Waves 2 and 3. It is interesting that subjective perceptions of job security were quite stable as compared to relatively unstable objective layoff contact experiences. When discussing findings regarding our proposed alternative hypotheses, then, it is important to keep in mind that the core finding is stability in our outcome variable: problem alcohol use at Wave 2 or Wave 3 is predominantly a function of problem use in previous waves. Any standard, reverse, or reciprocal effects over and above this stability are likely to be modest. However, given the limited amount of variance left to explain, any effect that emerges might be viewed as substantively important as it identifies a possible mechanism to modify otherwise very “set” behaviors, beliefs, or perceptions. Making comparisons. Tables 5 and 6 summarize fit indices for layoff contact and job security models respectively as well as results of Chi difference tests. Tables 5 & 6 about here All models show very good fit with CMIN well under 3.0, TLI and CFI over .90, and RMSEA at or under the .05 threshold. The important question, however, is which model (alternative hypothesis) best fits the data? Because our alternative models are nested models we can utilize the Chi difference test to assess if the loss of parsimony due to freeing paths in the more complex models (models with cross-lagged paths included) comes at too high a cost in fit (Bentler & Bonnett, 1980; Joreskog & Sorbom, 1993; Garson, 2001 ;Kline, 1998). A significantly different chi-square result suggests that improvement in fit over the baseline model justifies retaining the paths in the more complex model. Layoff contact. For the layoff contact models (Table 5), the only comparison that shows a statistically significant difference is the stability model (L1) and the reverse hypothesis model DOWNSIZING AND ALCOHOL USE 21 (L3) comparison (Chi2 diff L1-L3 (4) = 14, p < .01). These results suggest that problem alcohol use influences subsequent escape reasons for drinking and escape reasons for drinking influence subsequent layoff contact experiences. When looking at path coefficients for model L3 (Table 3), however, we note that only two of the four path coefficients are significant: the path from alcohol problems Wave 1 (Alcprob1) to escape reasons Wave 2 (Escape2) at .20, p < .002 and from escape reasons Wave 2 (Escape2) to layoff contact Wave 3 (Layoff3) at .11, p < .041. When we drop the non-significant paths to create a more parsimonious model (L5) and compare this model to the stability model (L1) we note that there is still a significant difference between the two models (Chi2 L1-L5 diff (2) = 13, p < .01). Further, the difference between this model (L5) and the original reverse hypothesis model (L3) is not significant. These results suggest that the L5 model fits the data as well as the less parsimonious model L3, while still showing substantive benefit by including two reverse paths in addition to the stability paths. The L5 model, with its two reverse paths, represents the “best” model of the layoff contact variables as in Figure 3 below. FIGURE 3 about here This model partially supports a reverse rather than standard causal hypothesis regarding the relationships between layoff contacts, escape reasons for drinking, and problem alcohol use. Specifically, we can say that over and above stability coefficients and within wave relationships, problem alcohol use at Wave 1 has a modest, significant effect on change in escape reasons for drinking such that an increase in problem alcohol use at Wave 1 leads to an increase in the tendency to use alcohol to escape at Wave 2. Similarly, over and above stability and within wave relationships, escape reasons for alcohol use at Wave 2 has a small, but significant effect DOWNSIZING AND ALCOHOL USE 22 on change in layoff contact such that an increase in escape reasons for drinking causes a tendency toward more direct forms of layoff contact at Wave 3. Job security. We see a slight variation on this pattern in the job security model comparisons (Table 6). We note that the only significant model comparison is between the stability model (J1) and the reverse model (J3) with Chi diff (4) = 10, p < .05. This suggests that the improvement in fit justifies the addition of reverse paths. However, as we noted for the layoff contact comparisons, a closer examination of path coefficients (TABLE 4) indicates that only one of the additional reverse paths, the path from problem alcohol use in Wave 1 to escape reasons Wave 2, is statistically significant. When we drop the three non-significant paths, and compare model J5 to the stability model, we note that there is still a significant difference between this more limited reverse causal model and the stability model (Chi2 diff J1-J5 (1) = 8, p < .01). This more parsimonious model has equivalent fit to the more complex reverse model with four paths so we retain this model as “best” for job security. This model partially replicates the findings from the layoff contact model in that there appears to be a fairly strong reverse relationship between problem alcohol use at Wave 1 and escape reasons for drinking at Wave 2. There do not appear to be any additional causal relationship with job security, however. As compared to layoff contact, there are fairly high stability coefficients for independent variable of perceived job security. This means there are not significant amounts of variance left to explain in job security, escape reasons for drinking, and alcohol use at Wave 3 once we account for stability and within wave associations. Discussion In this study we are interested in exploring why empirical relationships between workrelated stressors (specifically repeated downsizing experiences) and problem alcohol use are DOWNSIZING AND ALCOHOL USE 23 generally either weak or inconclusive. Prevailing theory regarding these relationships suggest that negative workplace experiences should “spillover” into non-work domains and stimulate the use, and possibly abuse, of alcohol as a coping device. Because most studies in this domain have been conducted using cross-sectional or incomplete two-wave panel methods, testing this “standard” causal linkage has not been feasible. By systematically testing alternative causal hypotheses using fully cross-lagged three-wave mediational models, we attempt to clarify what have to date been largely null findings regarding the relationship between problem alcohol use and the workplace. Our results suggest that one reason for weak support for spillover or stress-reduction models of alcohol use is that problem alcohol use appears to be very stable. That is, the primary causal driver of problem alcohol use at Wave 2 or Wave 3 is the presence of problem alcohol use at Wave 1. Once this stability is partialled out, there is little variance left to explain by workplace (or other) factors. Apart from this primary finding, our results do suggest that a reverse causal model may be plausible when examining the relationship between alcohol problems and the workplace. The only significant paths in the models apart from the stability coefficients are paths from problem alcohol use at Wave 1 to escape motivations at Wave 2 as well as a significant direct path in the layoff contact model from escape reasons at Wave 2 to layoff contact at Wave 3. We did not see any evidence for the spill-over hypothesis, but did see some evidence that problem alcohol use might lead to subsequent passive coping responses and, perhaps, increase the chance of being more proximal to more personally-relevant layoff experiences. In sum, these data suggest that alcohol use may be antecedent to not consequence of