2.1. Method
In order to validate the three clinical structures obtained (cognitive, somatic and affective), the parameters of the BLIM have been estimated for each of the three structures based on data from the 383 participants.
Participants: The sample included: a clinical group consisted of 38 subjects with MDE
(who were diagnosed with major depressive disorder or bipolar disorder) of the Neurosciences, Mental Health, and Sensory Organ (NESMOS) Department of La Sapienza University, Rome. (This sample is the same of the previous research, chapter 5). A non-clinical sample of 345 individuals recruited in the area of the University of Padova (both students and non-students; 68% were female). The majority of participants had a high school diploma, and their ages ranged between 19 and 58 years. (The criterion of exclusion was the same of the previous research, chapter 5).
Administration: All the research participants completed informed consent and
sociodemographic forms before answering the questionnaire items. No time limit was imposed to complete the questionnaires. All 383 subjects completed the written form of QuEDS (41 dichotomous items). This tool has been described in the previous research, presented in Chapter 5. At clinical intake, participants provided written, informed consent for potential research analysis and anonymous reporting of clinical findings in aggregate form, in accord with Italian legal and ethical requirements. The study was conducted in accordance with the Declaration of Helsinki. All participants entered the study of their own free will and they were informed in detail about the aims of the study, the voluntary nature of their participation, and their right to withdraw from the
study at any time. Furthermore, participants could ask for restitution about their own score, providing authors with their own auto generated code, used during the administration phase.
Procedure: The estimate was performed with a specific version of the Expectation-
Maximization Algorithm (Dempster, Laird, & Rubin, 1977) for MatLab, i.e., CEMBLIM. For the description of the algorithm, refer to Spoto (2011). The fit of each of the three models has been tested by Pearson’s chi-square. It is well known that for large data matrices (as those used in the present study) the asymptotic distribution of χ2
is not reliable. Therefore, a p-value for χ2 has been obtained by parametric bootstrap (n.
of replications = 5000).
2.2. Results and discussion
In Table 1 are displayed the fit indexes of the three models estimated by CEMBLIM. Results show adequate fit indexes for all three models. In particular, this table shows the global fit indexes obtained for the three models together with the corresponding p- values obtained by parametric bootstrap and the number of clinical states for each sub- scale.
Table 6.1:The global fit indexes of the three models Sub-
factor
Num. of states df χ 2 Bootstrap P
Cognitive Somatic Affective 124 163 142 32144 15972 3928 23348 7237 8696 .07 .16 .06
Table 6.2, displays the false positive (η) and a false negative (β) for each item of the three sub-scales of QuEDS. Note that the sum of η and β do not exceed the value 1 in line with the BLIM model assumptions.
Formally: η + β < 1 for all q ∈ η < 1- β and β < 1- η
Indeed, the probability that a false positive (η) occurs must be less than the complement to 1 of the probability of a false negative (β). In other words, the probability of a false positive (η) on an item i, must be less than the answer is “yes” without false positive. Same explanation as regards the false negative.
Table 6.2: Estimated parameters β and η for each item of the three sub-scales
COGNITIVE SOMATIC AFFECTIVE
β η β η β η
Item 5 0.25 <.001 Item 1 0.20 0.04 Item 7 0.44 0.02 Item 6 0.18 <.001 Item 2 <.001 0.01 Item 8 0.14 <.001 Item 9 0.50 <.001 Item 3 0.13 0.04 Item 12 0.09 <.001 Item 10 0.09 <.001 Item 4 0.01 <.001 Item 15 0.11 0.01 Item 14 0.33 0.01 Item 11 <.001 <.001 Item 17 0.45 0.07 Item 19 <.001 0.03 Item 13 0.15 <.001 Item 18 0.08 0.03 Item 20 0.19 <.001 Item 16 0.12 0.01 Item 29 <.001 <.001 Item 21 <.001 <.001 Item 22 0.10 <.001 Item 34 0.06 0.07 Item 24 <.001 <.001 Item 23 0.05 <.001 Item 36 0.36 0.02 Item 25 0.03 0.01 Item 26 0.02 0.04 Item 37 0.17 0.01 Item 27 <.001 0.18 Item 28 0.06 <.001 Item 38 0.09 <.001 Item 30 0.44 0.01 Item 31 0.12 <.001 Item 40 <.001 <.001 Item 32 0.11 0.06 Item 35 0.16 0.07
Item 33 0.02 <.001 Item 39 <.001 <.001 Item 41 0.01 <.001
As we can see from the table, the single items’ η parameters seem quite small for almost all items. On the contrary, there are two items with high probability of false negative (β)
in the cognitive scale of QuEDS: item 9 (β = .50) “I think my life is hell and I only deserve to feel bad”, and item 30 (β = .44) “I feel too much on the other people that it would be better if I killed myself”. The false negative indicates that some subjects answered “no”, although the symptoms investigated by that item were also investigated by other items where individuals answered “yes”. This could mean that the interpretation of these specific items could be complicated for some subjects (maybe because they are composed of two sentences) and therefore the symptoms included in those items were more easily understood within other items. To better understand the links among the items and the attributes look at Table 5.5, Chapter 5.
In the affective scale of QuEDS, there are two other items with high false negative rate: item 7 (β = .44) “I keep crying very easily” and item 17 (β = .45) “I often feel like crying, but I cannot do it”. Both high values of β are related to “crying”; in this case we could suggest that the feel like crying as well as the “crying” were underestimated in the non-clinical sample. We can suppose that the subject could either intentionally fake the specific answer, or the subject’s answer could be affected by the poor introspection about “crying”. However, all the other items show reasonable error parameters.
Parameter estimates will be essential for the implementation of the algorithm to obtain the adaptive form of the QuEDS.