First Simple makes last complex: construct irrelevant variance effects in the test of grammatical comprehension

Megan Esler, Maria Garraffa

Research output: Contribution to conferencePosterpeer-review

Abstract

Reliability of a test increases with test length, offering opportunities for insights on a phenomenon. However ‘artificial inflation’ may pose psychometric disadvantages to item score. An extraneous variable (a construct Irrelevant Variance) can negatively affect assessment outcome, with items towards the end of a test receive less attention.
The current study draws upon a widely used test for sentence comprehension; the Test for Reception of Grammar (TROG-2) (Bishop, 2003). Specifically, when looking at standardised scores, the normative data illustrated as developmental trend graphs depict age equivalent scores for each construct, and appear to follow an incremental format. However, in the final constructs, participants at the higher end of the age scale, 14 to 16y display a low performance. This study examines whether the order of the constructs or ‘blocks’, although representationally coherent in an incremental complexity scale, can affect TROG-2 score. TROG-2 is suitable for children from 4 years old to adults. It has been widely used across various clinical groups (hearing impairment, SLI, Neurological patients).

Methods: 40 participants (18 male) were randomly allocated in two groups. All were of undergraduate education and English native speakers. Prior to being tested with TROG-2, each participant completed a Digit Span test to establish equal working memory in both group and a minimum inclusion score of 4 for each participant.
One half of the sample, Group A, were in the control condition whereby they were tested using TROG-2 in the correct order, whilst remaining half, Group B, were presented with the blocks in TROG-2 in the reverse order.

Results: Group A (N=20) was associated with a TROG-2 test score of M=16.35 (SD=2.35), Group B (N=20) who completed the study in the reverse order, was associated with a numerically overall higher test score of M=18.15 (SD=1.23). The difference in TROG-2 overall test scores between the two groups was significant, t(28.66)=-3.04, p=.005.
Overall TROG-2 scores were calculated on a pass/fail basis per block. In order to pass a block, all four items within a block must be successfully answered. Group B scored higher in all blocks individually excluding Block A, B, C, D, L, N (see Table 1 and Figure 1). Block T scores for Group A (mean rank=14.53) and Group B (mean rank=26.48) were significantly different (p=.001).

Discussion: The aim of this study was to determine whether the directionality of which the constructs or blocks are presented to participants would affect their block score and overall score on TROG-2, assessing whether the low standardised scores reported in the final blocks can be explained by task difficulty or whether it can be attributed to an order effect. The order manipulation was effective with participants in the reverse group making significant less error on the centre-embedded relative clauses. Data will be discussed in a more comprehensive model including the interaction with working memory reported in the literature for this specific linguistic construct.

Cite this