Page 41 - IJPS-11-6

International Journal of
Population Studies Gender gaps in reporting limitations

the frequencies of vignette reporting, whereas Table A2 For the vignette component analysis, we used a set of
presents the Pearson correlation coefficients among all the vignettes from the survey data. All respondents perceive
vignettes, showing low dependency among most of them. each particular vignette j to be consistent with a latent level
of work limitation, h , which is independent of the
v*
2.2.2. Covariates ij
respondent’s characteristics. This follows the assumption
To ensure robust results, we controlled for several key of vignette equivalence, which requires that all individuals
factors in our analyses, in line with common practice in the perceive the vignette as corresponding to a health level on
literature, that have been shown to play a confounding role the same scale. If this assumption does not hold, then one
in self-reported work disability (Kapteyn et al., 2009; Yin could not attribute variation in the rating of a given vignette
& Heiland, 2022). These factors include sociodemographic to reporting heterogeneity. There has been little formal
characteristics, health conditions, and employment status. testing of this assumption (Murray et al., 2003).
Sociodemographic variables include gender (women vs. Consequently, the latent work limitation of each vignette j
men), age (50 – 55, 56 – 60, 61 – 65, and 66 – 70 in years), as perceived by respondent i can be specified as an intercept
race (Non-Hispanic White, Non-Hispanic Black, Non- plus random measurement error:
Hispanic others, Hispanic), educational attainment (less
v
than high school, high school, some college such as junior h ij * v = α + ε ε, v ij ~ (0,1)N (I)
j
ij
college or college without degree, and college or above,
v
with high school as the reference group in estimation), and We normalize α to be zero, and ε independent of each
ij
1
present employment status (whether currently working other and of the covariates x . The respective observed
i
v
v*
for pay). We further controlled for 7 dichotomous health categorical rating h is related to h through the following
ij
ij
conditions: High blood pressure (or hypertension), mechanism:
diabetes (or high blood sugar), cancer (cancer or a v = kif µ h k −1 ≤ * v < h µ k = ,k … 1, ,5
malignant tumor or any kind except skin cancer), lung ij i ij i (II)
problem (chronic lung disease such as chronic bronchitis
1
5
2
or emphysema except asthma), heart problem (heart attack, with , and µ i 0 µ = −∞and 5 i = +∞ . The
i
i
i
coronary heart disease, angina, congestive heart failure, or exclusion restriction in Equation I allows us to identify the
other heart diseases), arthritis (arthritis or rheumatism), cut-points as functions of the respondents’ characteristics:
and obesity. In addition, depression symptoms, measured k k k k
… ,k
using the Center for Epidemiological Studies Depression µ i γ = 0 i γ+ i δ+ x = g 1, ,4 (III)
(CESD) Scale, and disability, as indicated by the number of where x is a vector of a respondent’s characteristics
limitations in activities of daily living, were also controlled (except gender); g is a gender dummy with women
i
for in the statistical modeling. Unless otherwise stated, the being one. The effect δ reflects the gender differentials
i
k
first category of non-dichotomous variables serves as the in response scales. A positive estimate of δ suggests that
k
reference group in the analysis.
women respondents tend to have a higher threshold than
2.3. Analytical strategies men in assessing a vignette character’s disability. That
is, women are stricter and likely rate a given disability
Our analysis is based on the hierarchical ordered probit vignette as less severe than men. Conversely, a negative
(HOPIT) model that is commonly used in vignette studies estimate of δ would suggest more inclusive criteria by
k
(Kapteyn et al., 2009; King et al., 2004), which uses vignette women in rating the work limitation vignettes compared
data to model the cut-points as functions of respondent to men.
characteristics. These cut-points are then applied to the
model for self-reported work limitations, allowing us to As in a standard ordered probit model, the second
distinguish between actual health effects and reporting component of the HOPIT defines the latent level of the
effects. respondent’s own work limitation and the process that links
this latent variable to the observed categorical variable. The
The HOPIT model has two components: The vignette
component, which captures reporting behavior by modeling difference is that the cut-points are no longer constant but
cut-points as functions of respondents’ characteristics can vary across respondents and are determined by the
(accounting for reporting heterogeneity), and the health vignette component of the model.
component, which represents the relationship between the Specifically, the second component of the HOPIT
respondent’s own work limitations and covariates, with the defines the latent level of individual own work limitation,
s*
cut-points determined by the vignette component. h , as:
i
Volume 11 Issue 6 (2025) 35 https://doi.org/10.36922/ijps.1969