ترس ها و تحقق ناامنی استخدام

We investigate the validity of subjective data on expectations of job loss and on the probability of re-employment consequent on job loss, by examining associations between expectations and realisations. We find that subjective expectations data reveal private information about subsequent realisations both of job loss and of subsequent re-employment. We also compare the use of verbal and numerical descriptors for subjective probability scales. As predictors of subsequent job loss, the expectations data perform better with numerical than with ordinal verbal descriptors. On average, employees overestimate the chance of losing their job; while they underestimate the difficulty of job replacement. We recommend that survey items on employment insecurity should be explicit about each risk under investigation, and utilise a cardinal probability scale with discrete numerical descriptors.

What can be learned from workers' expectations about employment and unemployment? In this paper we evaluate indicators of the subjective expectations underpinning employment insecurity. From studies in the US and the UK it is found that expectations of job loss predict subsequent unemployment, albeit with some error and showing a degree of pessimism (Stephens, 2004 and Campbell et al., 2007). However, the relation between subjective expectations and unemployment outturns could be spuriously positive, or conversely underestimated, if personality traits affect the chance of job loss and also the perception of the risk of job loss. We address this problem by estimating a model with fixed effects, with results similar to those in the literature, using data from Germany and Australia (countries with dissimilar labour market institutions). In all cases a rise in the perceived chance of job loss successfully predicts a rise in the probability of unemployment, but the marginal effect is considerably less than one. What is hitherto unknown is whether workers can also predict subsequent replacement conditional on job loss. We investigate this issue, and find that there is robust criterion validity in the data on workers' expectations of job replacement. In contrast to their expectations of job loss, workers' expectations of job replacement reveal evidence of unwarranted optimism. We also address whether the predictive power of job loss expectations data may be affected by the wording of the question, in particular the response scale. A cardinal scale on the risk of job loss, in which respondents are asked to state the probability of job loss, has considerable attractions for economists (Manski, 2004), but might not correspond to the categories individuals use when thinking about job loss probability. Some surveys instead deploy less precise ordinal descriptors (such as ‘very unlikely’, ‘unlikely’, ...), which rely on assuming respondents have a shared and internally consistent understanding of terms, and restricts comparisons to ordinal rankings. Yet there is hitherto no evidence as to which approach elicits the most information about expectations. Taking advantage of a decision to change the scale in an otherwise consistent series on expectations of jobs loss, we compare the predictive power of a cardinal scale with that of an ordinal scale. We find, not only that the cardinal scale has criterion validity, it also appears to do somewhat better than the ordinal scale. These matters appear especially relevant, given that direct subjective indicators of economic expectations are increasingly finding favour within economics as valuable tools for predicting well-being and behaviour (Manski, 2004). Earlier studies have examined and supported the validity of expectations data in various other domains (Hurd and McGarry, 2002, Smith et al., 2001, Hamermesh, 1985, Dominitz, 1998 and Manski, 2004). Attaining good measures of subjective employment insecurity would seem to be pressing, and not just for their predictions of subsequent labour market status. Perceived job insecurity and a lack of employability have been found to generate anxiety and substantially lower the well-being of workers and their dependents, to inhibit consumer spending, and to reduce wage growth (Wichert, 2002, Benito, 2004, Lusardi, 1998, Carroll et al., 2003, Campbell et al., 2007, Di Tella et al., 2003 and Green, 2011). The impact of insecurity accounts in part for the highly detrimental effects of aggregate unemployment on the average well-being of populations. Nevertheless, the effects of employment insecurity vary across socio-economic categories, and between countries with different employment protection institutions; and this variation highlights the need for the different expectations behind feelings of employment insecurity to be more precisely specified in survey questions (Clarke and Postel-Vinay, 2004, Green, 2006 and Green, 2009). Employment insecurity comprises multiple parts, involving the risk of job loss, the chances of not finding another job, loss of income while unemployed, and uncertainty over job content. Unambiguously phrased survey instruments to capture each of these elements of uncertainty are needed, together with tested protocols to allow respondents to represent their expectations as reliably as possible. Our focus in this paper is on the first two categories. The world-wide economic downturn in 2008 serves to add some urgency to this task. The paper proceeds as follows. The next section reviews measures of employment insecurity in some prominent national and international surveys, considers arguments and evidence about the use of probabilistic or ordinal scales when measuring expectations, and connects our study to investigations within psychology of pessimism and optimism. Section 3 describes the data, the main findings are presented in 4 and 5 concludes with a discussion of the implications for survey question design and for employment insurance.

Indicators of expectations of labour market agents are potentially valuable for analysing a wide range of labour market behaviour, and they are in particular at the heart of the measurement of employment insecurity. However, before their wide-spread adoption in labour market research is likely, it will be necessary for researchers to develop greater understanding of their validity. To capture employment insecurity there is a good case for using questions that separately capture perceptions of the risk of job loss, uncertainties about the work itself, and uncertainties about the consequences of job loss. To measure these perceptions one might opt to defend any format that successfully elicits expectations, whether or not the expressed expectations are realised. If workers say they have a high fear of job loss this might affect their behaviour even if that fear is not warranted. Normally, however, economists will find expressions about labour market expectations of most interest if they convey private information — that is, information about the respondents' particular circumstances or personal intentions that would not typically be collected in other ways by researchers. In this light, we have reported several new findings based on the two extant, nationally representative, panels that regularly gather direct information about labour market expectations, the German Socio-Economic Panel and the Household Income and Labour Dynamics in Australia Survey. First, the expectations of job loss questions in both cases are found to robustly predict the probability of subsequent job loss, thereby providing solid support for the value of such questions in labour market analysis. While similar findings had been reported for a sample of older male workers in the US (Stephens, 2004) and for two nationally representative samples in Britain (Campbell et al., 2007), we have now demonstrated this predictive power for the first time using a fixed-effects panel estimator, and in two further nationally representative samples for Germany and Australia. Taken together, there can now be some confidence in the use of such expectations questions and with regard to individuals' ability to provide useful information in their responses to questions regarding their expectations of future job loss. Our second finding is that employees' perceptions of re-employment probabilities in the event of job loss are also robust predictors of their outcomes. We believe that this adds further support to the utilisation of expectations data in employment research. Third, the analysis of expectations and realisations establishes that there is merit in using cardinal rather than ordinal scales. Since the meaning of numerical scale points is unambiguous while that of verbal descriptors might differ among respondents if their understanding of language is heterogeneous or if the words are vague, cardinal scales are in principle preferable. Moreover, cardinal scales offer analytical advantages, in that marginal changes in probability are commensurate along the scale, which is not true of ordinal verbal descriptors. Nevertheless, it was an open question as to whether the responses on cardinal scales in practice can capture valid representations of what workers expect, given that not all respondents can be assumed to have sufficient skills and forethought. We have compared goodness-of-fit measures in GSOEP from before and after the switch from ordinal to cardinal scales in our analysis of job loss, and found that the cardinal scales appear to perform better on these grounds. It would appear that individuals can systematically assess the likelihood of important events such as job loss and re-employment using numerical, probabilistic scales, and do so somewhat better than with ordinal verbal scales. Nevertheless, the numerical scales revealed a small minority of inconsistencies in the responses, showing imperfect understanding of the questions.28 This systematic evaluation of the relative merits and demerits of numeric response scales for expectations questions is a further contribution of this paper. Fourth, although the expectations are found to have to have predictive power, in both cases and in both countries there are significant biases. There is a pattern of unrealistic pessimism over job insecurity (over-estimation of the chance of job loss) for all workers who have any fear of job loss, while those who think that they have no chance at all of job loss underestimate the small risk they face. The average perceived chance of job loss is roughly twice the frequency of subsequent occurrence. The consistency in the degree of unrealistic pessimism taken across the different countries and datasets is quite remarkable. Conversely, individuals' expectations of re-employment tend to be optimistic on average, with the proportion of job-losing Australians being re-employed in equal or better jobs being 19 percentage points less than the average probability that those individuals had anticipated. While these biases do not invalidate the use of expectations data, they invite explanation, which should be the focus of further research. Also warranted is a consideration of the implications of pessimism and optimism for private insurance demand, and for other outcomes including consumer demand and well-being. Prima facie, there would seem to be contradictory impulses, with job insecurity pessimism implying over-insurance while unrealistic optimism suggests the opposite. In addition to these implications for studying further the links between expectations and economic behaviour, our findings contain implications for how survey designers should go about measuring expectations in the field of employment. Given the resurgence of unemployment around the world in the current era, we believe that there is a good case for measuring employment insecurity explicitly in major labour market surveys, and the findings here indicate that it is worth collecting direct measures of workers' subjective expectations. Unlike in many previous studies of job or employment insecurity, the specific employment event about which expectations are formed needs to be clearly formulated – as suggested by Manski (2004) – and there is a case also for extending the domains reviewed here to include uncertainties about work characteristics. Another further extension would be to introduce and validate questions addressing the confidence or ambiguity with which expectations are held. Our analyses of the existing data show that it is both viable and advisable to utilise numerical scales. The particular scale to be used should ideally allow for the non-linearity at the upper end of the scale shown in Fig. 2, which suggests that it would not be entirely reliable to treat a marginal change in job loss expectations as proportional to a marginal change in the objective incidence of job loss. The scale should also not attempt to capture unrealistic precision. Hence, instead of soliciting responses as continuous percentages, a set of discrete percentage points is preferable, though this has to be balanced against the extra time and space within the survey protocol.29 Extra variation can be allowed for around the well-populated parts of the distribution; hence, in the case of job loss risk an appropriate set of scale percentage points might be: 0, 5, 10, 15, 20, 30, 40, 50, 60, 70, 80, 90, 100; while in the case of the difficulty of re-employment, decile points would capture the very large proportion of the distribution.