We use a sample of z ∼ 3 Lyman-break galaxies (LBGs) to examine close pair clustering statistics in comparison to Λ cold dark matter (ΛCDM)-based models of structure formation. Samples are selected by matching the LBG number density, ng, and by matching the observed LBG 3D correlation function of LBGs over the two-halo term region. We show that ultraviolet (UV) luminosity abundance matching cannot reproduce the observed data, but if subhaloes are chosen to reproduce the observed clustering of LBGs we are able to reproduce the observed LBG pair fraction (Nc) defined as the average number of companions per galaxy. This model suggests an overabundance of LBGs by a factor of ∼5 over those observed, suggesting that only one in five haloes above a fixed mass hosts a galaxy with LBG-like UV luminosity detectable via LBG selection techniques. This overdensity is in agreement with the results of a Millennium 2 analysis and with the discrepancies noted by previous authors using different types of simulations. We find a total observable close pair fraction of 23 ± 0.6 per cent (17.7 ± 0.5 per cent) using a prototypical cylinder radius in our overdense fiducial model and 8.3 ± 0.5 per cent (5.6 ± 0.2 per cent) in an abundance matched model (impurity corrected). For the matched spectroscopic slit analysis, we find and 5.1 ± 0.2 (1.68 ± 0.02) per cent, the average number of companions observed serendipitously in randomly aligned spectroscopic slits, for fiducial slits (abundance matched), whereas the observed fraction of serendipitous spectroscopic close pairs is 4.7 ± 1.5 per cent using the full LBG sample and 7.1 ± 2.3 per cent for a subsample with higher signal-to-noise ratio. We conduct the same analysis on a sample of dark matter haloes from the Millennium 2 simulation and find similar results. From the results and an analysis of the observed LBG 2D correlation functions, we show that the standard method of halo assignment fails to reproduce the break, or up turn, in the LBG close pair behaviour at small scale (≲20 h−1 kpc physical). To reconcile these discrepancies we suggest that a plausible fraction of LBGs in close pairs with lower mass (higher density) than our sample experience interaction-induced enhanced star formation that boosts their luminosity sufficiently to be detected in observational sample but are not included in the abundance matched simulation sample.