We wish to investigate the measurement of borderline personality disorder (BPD) within the 11-12 year-old cohort of the ALSPAC data. Specifically, we intend to employ item response theory (IRT) methods on the BPD symptom criteria. IRT constitutes a latent trait approach to psychological measurement, modeling the probability of endorsing a given item (e.g., presence of affective instability) as a function of an individual's standing on the underlying construct (e.g., BPD).

The basic unit of IRT is the item characteristic curve (ICC), which depicts visually the relationship between the construct and a person's response to an item (Embretson & Reise, 2000). The shape of this curve is influenced by at least two parameters: difficulty and discrimination. The difficulty parameter represents the point at which the probability of endorsing an item is 50%. The discrimination parameter is the slope of the ICC at the value of the difficulty parameter. These parameters will provide important information for evaluating the performance of each BPD criterion, such as the level of BPD needed to meet the criterion and the degree to which it differentiates among higher and lower levels of BPD (Feske, Kirisci, Tarter, & Pilkonis, 2007).

Most applications of IRT require that the construct being measured is unidimensional. Several factor analytic investigations of the BPD criteria in adults have supported a unidimensional structure (Aggen, Neale, Roysamb, Reichborn-Kjennerud, & Kendler, 2009; Feske et al., 2007; Fossati, Maffei, Bagnato, Donati, Namia, & Novella, 1999), while others have reported a structure sufficiently unidimensional for IRT purposes (Johansen, Karterud, Pedersen, Gude, & Falkum, 2004; Sanislow et al., 2002), as evidenced by high factor intercorrelations (Embretson & Reise, 2000). To date, only one study (Becker, McGlashan, & Grilo, 2006) has conducted factor analysis of the BPD criteria in a youth sample, the results of which did not reveal a unidimensional structure. However, this study employed principle component analysis with orthogonal rotation and, thus, did not report the magnitude of factor intercorrelations. Taken together, these studies provide reason to believe that the BPD criteria may be sufficiently unidimensional to employ a traditional IRT-based model. However, in the event that the BPD criteria are not unidimensional in the ALSPAC data, multidimensional or categorical confirmatory factor analysis will be employed.

In order to investigate our aims, we will need the CI-BPD data that correspond to each of the 9 DSM-IV symptom criteria, including the questions used to inform judgment of whether the symptom was present.