Page 79 - Contributed Paper Session (CPS) - Volume 7

CPS2033 Ronnie P.
2. Methodology
To study misclassification of the aforementioned type, this paper uses
formal mathematical proofs, simulations and an empirical analysis.
In the empirical analysis, a register-based evaluation of vocational
rehabilitation, serves as an illustration. As commissioned by the Swedish
government in 2011, the Public Employment Service (PES) and the Swedish
Social Insurance Agency (SIA) implemented in 2012 a model involving
enhanced cooperation between the two government agencies. The aim was
to target individuals entering sick leave and identify the need of support in
order to regain work ability.
The individuals' work-ability, both from a medical and a labour market
viewpoint, was identified through a joint assessment meeting (JAM) where
individuals during an evaluation period had meetings with PES and SIA.
Because individuals were not randomly assigned to JAM, but instead assigned
according to the decision of each individual's case-worker there is reason to
believe that the characteristics of individuals assigned to JAM are not similar
to those not assigned to JAM.
Thus, when evaluating the causal effect of being called to JAM, on for
instance the total extent of sick leave, it is necessary to make the JAM group
and non-JAM group comparable. To handle the selection bias, Fowler et al.
(2017) (which was funded by Swedish government) designed a protocol for
the estimation of the causal effect of JAM. The protocol includes several
confounders, for instance year of birth, sex, country of origin, marital status,
education level.
Furthermore, the two confounders, Mental and behavioural disorders
(Chapter V in the ICD-10 classification) and Diseases of the musculoskeletal
system and connective tissue (Chapter XVIII in the ICD-10 classification), were
pointed out as extra important. Following the protocol, the effect of JAM on
sick leave was evaluated in Fowler et al (2017).
However, although different sources of biases are acknowledged in Fowler
et al. (2017), misclassification of Mental and behavioural disorders and
Diseases of the musculoskeletal system and connective tissue is not
mentioned. Yet, studies show that there is reason to believe that for instance
mental disorders are underreported (e.g. Ezzat 2015; Takaynagi, 2014). It
would therefore be important to study how misclassification affects the causal
effect estimates in Fowler et al. (2017).
Moreover, because the degree of one-sided misclassification probably
differs when comparing Mental and behavioural disorders and Diseases of the
musculoskeletal system and connective tissue this empirical example would
further lend itself to illustrate the bias of misclassification.

66 | I S I W S C 2 0 1 9