The analysis is based on annual time series data from the EU-SILC study from 2004
to 2011, which are publicly available 21]. The dependent variable was the percentage of people who had medical needs but did
not use healthcare services. Independent variables were time [coded as 0:2004–2007
(no crisis), 1:2008–2009 (crisis starting, no austerity measures) and 2:2010–2011
(crisis with austerity measures in effect), median income, and unemployment. Initially
we used the augmented Dickey-Fuller test to see if the variables were stationary.
Given that the variables presented unit roots, we tested if the residuals of the ordinary
least squares (OLS) model presented unit roots (without constant and trend). Since
testing confirmed that regression residuals had no unit roots, regression was judged
as non-spurious, and the variables co-integrated. A final model choice was based on
the information criteria in AIC and BIC. We also tested for normality and heteroskedasticity
of the residuals with the skewness and kurtosis test and the Breusch-Pagan/Cook-Weisberg
test, as well as performed a link test to check for specification error. Finally,
we tested for autocorrelation of standardized residuals via Durbin’s alternative test.
To study the effect of the financial crisis, controlling for other socioeconomic variables
on the outcome of interest, namely the reason for unmet healthcare needs, we merged
data from the two national surveys from 2006 to 2011 that were conducted by the National
School of Public Health 22]. The sample sizes were 4003 in 2006 (n
2006
?=?4003), and 6569 in 2011 (n
2011
?=?6569, n
total 2006
2011
?=?10,572), and they were both selected randomly based on stratification according
to prefecture (based on the residence of the respondents), degree of urbanity based
on NUTS II, age, and gender. Subjects were asked to report on experiences during the
preceding year. Both surveys used a common questionnaire based on World Health Organization
methodology 23] that had been validated in the past, and data collection involved a personal interview.
In 2006 the interviews were conducted in the home of the respondents whereas in 2011
the interviews were conducted by telephone.
The final sample that we used for the analysis included 3120 patients who reported
unmet needs. Specifically, 1243 of them were men (39.84 %), and the remaining 1877
were women (60.16 %), while the median age was 45 years. Among those who had stated
that they have a medical need, 894 respondents (28.65 %) identified financial reason
for not seeking care, whereas 2226 individuals (71.35 %) identified another reason
for unmet healthcare needs. More details for the final sample used in the analysis
are presented in Table 1.
Table 1. Distribution of unmet healthcare needs due to financial reasons per year
The analysis focused on those participants who reported a medical or healthcare need,
but no healthcare utilization. The outcome was dichotomized to 1 for unmet healthcare
needs due to financial reasons, and to 0 for unmet healthcare needs due to other reasons.
The final sample size was n
2006
2011
?=?3120 (n
2006
?=?1259, n
2011
?=?1861). We gained permission to access this dataset from the Department of Health
Economics, National School of Public Health.
Continuous variables were used as such, and Helmert coding was used for ordered variables,
including education and income level. Various dummy variables were created for the
nominal variables of employment and prefecture.
Statistical analysis was carried out in STATA 9.0. We used multiple logistic regression
(MLR) to assess the effect of the main variable (year of participation) on the outcome
(reason for unmet healthcare needs) controlling for various potential predictors or
confounders. Potential predictors (independent variables) in the model were the following:
a) gender (1: female, 2: male); b) age; c) self-reported health status (1: very bad,
2: bad, 3: medium, 4: good, 5: very good); d) existence of chronic health condition
(1: no, 2: yes); e) education level (1: no education, 2: elementary school, 3: high
school, 4: post high school and/or technical vocational education, 5: higher education,
6: university, 7: post-graduate education); f) income level (1: no income, 2: 1–500€,
3: 501–1000€, 4: 1001–1500€, 5: 1501–2000€, 6: 2001–3000€ and 7: 3001€+); g) employment
status (1: working, 2: unemployed, 3: retiree, 4: homemaker 5: student or soldier,
6: other); h) public social security health insurance (1: yes, 2: no); i) private
health insurance (1: yes, 2: no); j) urbanity status of permanent residence (1: rural,
2: urban); k) geographic prefecture (1: Attica, 2: East Macedonia and Thrace, 3: West
Macedonia, 4: Central Macedonia, 5: Epirus, 6: Thessaly, 7: West Greece, 8: Central
Greece, 9: Islands of Northern Aegean, 10: Islands of Southern Aegean, 11: Peloponnese,
12: Ionian Islands, 13: Crete); and l) year of survey (0: 2006, 1:2011). The appropriateness
and fit of the final models were checked using several diagnostic methods, such as:
i) link test, to test if the model suffers from specification error; ii) Hosmer and
Leme show goodness of fit criterion; iii) skewness and kurtosis test of normality
of the deviance residuals; and, iv) Brown and Forsythe test for the homoskedacity
of the deviance residuals. ROC curves were fitted to explore the interpretation value
of the models.