Flexible Work Options and Mothers' Perceptions of Career Harm

Abstract

Whether or not mothers, who often struggle with balancing work and parenting responsibilities, perceive that they face career harm in exchange for control over flexible work options at their jobs is an unanswered question. Using 2009 original data from a random-digit-dial telephone survey of 441 mothers located across the United States, this study focuses on how control over two latent variables measuring flexibility, flexible work arrangements (such as scheduling and place of work) and time-off options, influences mothers' career harm perceptions in a total of three work domains: (1) wages/earnings, (2) raises or promotions, and (3) job evaluations. We find perceptions of career harm among only one-fifth of mothers; in addition, control over time-off options reduced perceived career damage related to parenting duties. Mothers may have less to fear than previously hypothesized about the potential sacrifices they have to make when they have significant control over certain flexibility options.

ACKNOWLEDGMENTS

The author would like to thank Patricia Sheffield, Dr. Marc Weiner, and M.B. Crowley for their assistance with this manuscript. In addition, the Alfred P. Sloan Foundation provided generous financial support.

NOTES

Notes

¹ Interestingly, Schwartz later modified her strong views regarding Mommy Track costs. See Wood (2013:263).

² Caregiving here relates solely to child, and not elder care responsibilities.

³ We used the American Association for Public Opinion Research's (AAPOR) COOP3 cooperation rate and RR3 response rate definitions in our calculations. See AAPOR (2008). The cooperation rate is the proportion of respondents interviewed relative to the number of eligible respondents contacted. The response rate is the proportion of respondents interviewed relative to the number of eligible respondents in the sample. The response rate reported here is considered within the range of industry standards. See Kohut et al. (2012) and Pew (2004). We did not observe any systematic patterns of nonresponse bias, and our sample of working mothers is representative of working mothers nationwide (for additional information on these latter points, please contact the authors).

⁴ If our sample included other sociodemographic groups, we could make comparisons related to harm perceptions among them. See the Discussion and Conclusions section for a further explanation of our sample and its limitations.

⁵ Imputing values for categorical measures of a sample's social and demographic features is now quite common in both the social and medical sciences. For example, multinomial variables can be imputed using the multinomial logit link with passive imputation of the category-specific dummy variables. Ordinal variables and count variables can be imputed using an ordinal logit link. Other variables can be imputed using their natural links and exponential distribution families: logit link for binary variables and identity link for continuous variables. For more on the specifics of imputation, see Royston (2004) and White, Royston, and Wood (2011).

⁶ In addition, the quality of the measurement model is quantified by the composite reliability index (Raykov 1997), which plays the role of and has an interpretation similar to the coefficient of determination in regression analysis. In simple terms, it is the overall fraction of the item variance explained by the latent factor. High-quality, carefully designed specialized psychometric instruments may achieve a composite reliability of .8 to .9. In the context of survey research, however, lower values of .5 to .7 are more typical (Alwin and Krosnick 1991).

⁷ More precisely, to define a professional job, we used the List of Standard Occupational Classifications (SOC) as codes for jobs rather than industries. The List of SOC is defined by the Department of Labor, Bureau of Labor Statistics, and Occupational Employment Statistics. We then defined a professional job as any one listed that requires an academic degree, advanced certification, or some other type of formal training.

⁸ A drawback of factor regression is that it is known to produce biased estimates for the latent factor coefficients because of unaccounted measurement error. See Skrondal and Laake (2001).

⁹ With respect to our assumption that the three types of discrimination are equally important and thus can be added into one dependent variable, this was both a theoretical and methodological decision. The issue of estimating the importance weights of each component in a variable is typically handled on a case-by-case basis (Bentler 2004; Bollen, Glanville, and Stecklov 2007; Sijtsma 2009). With the relatively rare perceived discrimination events at issue here, applying unit weights appeared to be a sound approach. In addition, in preliminary analyses, we utilized logistic regression models for each perceived discrimination type and logistic regression for any type of perceived discrimination. Compared with all of these models, the negative binomial method that treated all perceived career impediments as equally important in one dependent variable produced the most parsimonious and easily interpretable results.

¹⁰ Allison (2002) has suggested that multiple imputation standard errors may be biased when VIFs exceed 2.5. The two highest VIFs were 2.08 for flexible work arrangements and 2.01 for time-off options.

¹¹ The overall F-statistics at the bottom of Table 3 were computed as:

where are parameter estimates from the negative binomial regression, k is the number of parameters (i.e., number of explanatory variables), and is their estimated variance–covariance matrix obtained using Rubin's rules. The numerator degrees of freedom of this F-distribution is the number of parameters (i.e., explanatory variables in the regression), and the denominator degrees of freedom is the minimum MI degrees of freedom (Reiter 2007).

¹² Note that since the standard deviation in this case for the factor is less than 1 at .927, we calculate the rate of the decline in perceived career harm experiences as 1 – exp(*[standard deviation of the factor]), where is the negative binomial regression coefficient.