You are on page 1of 2

PROBLEM 5:

The data are part of a real study and their use is allowed only for the learning process (which includes
this topic). Consider the Eviews file "diabetes_study" in which you have the following variables,
obtained for a sample of diabetic patients from Cluj county (data were collected in 2008):

COST_DIRECT_Y - the direct cost of this chronic illness, estimated for a patient (eg, by estimating
medication costs, hospitalization costs, regular medical visit costs, temporary or permanent loss of
employment, job changes, reduced hours due to illness, etc.)

COST_INDIRECT_Y - the indirect cost generated by this chronic disease, for each patient, over a
period of one year (eg the cost of time that the family and others spend with the patient both during
hospital treatment and during accompanying him to medical visits does not involve hospitalization,
loss of employment, reduction of working time and / or remuneration by family members or
caregivers, remuneration paid by the patient or his family to a professional health care provider, etc.)

The dependent variable is: TOTAL_COST_Y = COST_DIRECT_Y + COST_INDIRECT_Y.

..................

SOCIODEMOGRAPHICS CHARACTERISTICS (the names of the variables have been left in English;
explanations, where needed, will be given in class)

CLINICAL CHARACTERISTICS (the names of the variables have been left in English; explanations,
where needed, will be given in class)

A. Are you trying to arrive at one or more regression specifications, using the "general to specific"
strategy to show which factors (among those for which we have data) the total cost of diabetes
depends? Use a logY -> X specification. Give explanations (a priori, based on theory / logic) for
entering certain variables in the initial specifications, what are the expected signs and why.

Note: It is well known that too many variables in a regression can be a burden on data and increase
the risk of multicollinearity. Therefore, but without being mandatory, you can combine the "dummy"
variables of the same categorical variables in a way that has theoretical logic.

Ex: you can combine for example


MS_MARRIED+MS_COHABITATION Versus

MS_DIVORCED_SEPARATED+MS_NEVER_MARRIED+MS_WIDOWED.

The logic behind it is that those patients who have a life partner (either husband / wife, a concubine /
concubine) may be different in terms of cost induced from those patients who do not have a life
partner.

B. What are the significant variables and at what level of significance in the final specification /
specifications? Do you have explanations for the signs of all significant coefficients? But compared to
your initial expectations on the signs? Take a heteroskedasticity test, make the decision and take all
the steps required by your heteroskedasticity decision.

C. Often in the literature of the economic costs induced by chronic diseases such as diabetes it is
assumed that depression (which is a co-morbidity) influences the direct and indirect costs of the
underlying disease, increasing both, most often from the trend. depressives to overuse the system
(or to underuse it for a while until they get medical complications that increase exponential costs
from then on). Check the hypothesis of increasing the total cost of diabetes with the severity of the
depression, based on the data provided to you.

Note: It is known that "self-assessment of the severity of chronic disease" is often correlated with
depression. Maybe (but not necessarily) you want to consider only one of these two categorical
variables between which there could be a strong relationship. If necessary, perform a redundancy
test to rule out the patient's self-assessment of the severity of the disease and make a decision.
Interpret the results.

Or maybe, as a robustness test, you can offer more regression specifications in which some
depression coefficients remain at the same statistical significance (in all specifications provided as
final by you)?

You might also like