The $p$-ary code associated with the incidence structure of points and $t$-spaces in a projective space $\mathrm{PG}(m,q)$, where $q=p^h$, is the $\mathbb{F}_p$-subspace generated by the incidence vectors of the blocks of this design. The dual of this code consists of all vectors orthogonal to every codeword of the original code. In contrast to the codes derived from point-subspace incidences, the minimum weight of the corresponding dual codes is generally unknown, which makes the problem more challenging. In 2008 Lavrauw, Storme and Van de Voorde proved the following reduction: the minimum weight of the dual of the code derived from point and $t$-space incidences in $\mathrm{PG}(m,q)$ is the same as the minimum weight of the dual of the code derived from point and line incidences in $\mathrm{PG}(m-t+1,q)$. After a series of works by Delsarte (1970), Assmus and Key (1992), Calkin, Key and De Resmini (1999), the best known lower bound for the case of point-line incidences was established in [B. Bagchi and P. Inamdar: Projective geometric codes, J. Combin. Theory Ser. A, 99(1) (2002), 128-142]. The problem of determining the minimum weight of these codes admits a natural geometric interpretation in terms of multisets of points in a projective space which meet each line in $0$ modulo $p$ points. In this paper, by adopting this geometrical perspective and exploiting certain polynomial techniques from [S. Ball, A. Blokhuis, A. Gács, P. Sziklai, Zs. Weiner: On linear codes whose weights and length have a common divisor, Adv. Math., 211 (2007), 94-104], we prove a substantial improvement of the Bagchi-Inamdar bound in the case where $h>1$ and $m, p >2$.