THE NUANCES OF STUDYING FEMALE CANDIDATES’ ELECTORAL FORTUNES UNDER THE REGIME OF GENDER QUOTAS: THE UNDEBATABLE AND THE DEBATABLE

In their study, Jankowski and Marcinkiewicz claim to make two contributions. In the first part of their article, they criticize previous studies attempting to explain the assignment of ballot positions in OLPR systems, especially our study of the Polish case (Górecki and Kukołowicz Reference Górecki and Kukołowicz2014). Briefly speaking, they argue that our study incorrectly identifies the core outcome variable, which results in overemphasizing the adverse consequences of the introduction of quotas for female candidates. In particular, they claim that while our study acknowledges the fact that following the introduction of quotas the vast majority of female candidates were placed low on party lists, it misses the fact that the proportion of women among the candidates placed relatively high increased as well. This criticism is fair and, in my view, virtually undebatable. Arguably, the alternative analytical approach proposed by Jankowski and Marcinkiewicz constitutes a valuable contribution to the study of elections.Footnote ¹

The second part of Jankowski and Marcinkiewicz's study is their analysis of candidates’ electoral performance, focusing on explaining party-list vote shares obtained by candidates. The analysis is followed by a conclusion that candidate gender effects, and therefore the “paradoxical” effects of quotas, even if often discernible, are of so small a magnitude that they might have “occurred simply by chance” (Jankowski and Marcinkiewicz Reference Jankowski and Marcinkiewicz2019). Therefore, their argument follows, the “paradoxical” effects are practically negligible. I argue, however, that Jankowski and Marcinkiewicz's analysis cannot be interpreted as falsifying our claims. It is so for at least two reasons, which I explicate next.

First, unlike as in our study, Jankowski and Marcinkiewicz attempt to explain candidates’ party-list vote shares rather than the raw numbers of votes cast for those candidates. Unfortunately, in stark contrast to how they proceed with respect to the aforementioned operationalization of candidate ranking—for which they carefully explicate the limitations of our analytical approach—they do not at all embark on a criticism of raw numbers of votes as a measure of candidates’ electoral success. As a result, they do not effectively justify their abandonment of that measure in favor of candidates’ party-list vote shares. It can obviously be understood, to some extent, especially in light of the fact that the approach they chose is commonly pursued in the relevant literature. On the other hand, a critical reanalysis of previous findings should perhaps explicitly confront the potential consequences of choosing alternative operationalizations of the key concepts studied, including a reflection on the pros and cons of the chosen measure. The absence of such a reflection may be interpreted as implying that the choice of a particular operationalization is obvious and undebatable.

As long as candidates’ party-list vote shares as a measure of electoral success are considered, some “debatable spots” can be readily identified, however. Seen from the narrower viewpoint of the Polish case, certain limitations of that measure originate from the fact that in the Polish OLPR system populations of districts, and thereby the numbers of seats contested in each district, vary significantly. Jankowski and Marcinkiewicz's operationalization of electoral success results in a situation where the aforementioned variation is, so to speak, “conditioned out” and thus entirely disregarded. From the methodological viewpoint, disregarding some part of a potentially relevant variation and focusing on the remainder of it is a major decision that calls for justification.

Unfortunately, there are further, more general disadvantages of the measure chosen by Jankowski and Marcinkiewicz, ones that could, again, lead to genuine concerns. Namely, their operationalization of a candidate's electoral success results in equating the (large) electorates of large parties with the smaller ones. When calculating party-list vote shares, the only thing that is actually taken into account is a candidate's performance relative to his or her district-level running mates. The problem with this approach is that it disregards the differences in the district-level performance of parties, resulting in the corresponding differences in the numbers of seats won by parties. For instance, if one wants to win a parliamentary seat, then it might be better to be a candidate obtaining 10% of votes cast for a list winning five seats in a district than to be a candidate obtaining 20% of votes cast for a list winning just one seat. Hence, Jankowski and Marcinkiewicz's operationalization of electoral success constitutes an arguably distant approximation of the candidates’ actual ability to win seats. It is important to note here that, crude as it might seem on the surface, the operationalization emphasizing raw vote numbers tends to capture the aforementioned dynamics more accurately: candidates from parties “stronger” in a given district obtain larger numbers of votes.

Last but not least, if one wants to explain the behavioral aspects of voting under the regime of gender quotas and candidates’ ability to attract popular support, then all votes, and thereby all voters, should be treated equally. Thus, focusing explicitly on raw numbers of votes (as we do in our studyFootnote ²) has clear advantages, even if still being an arguably imperfect approach. All in all, both our measure and the one chosen by Jankowski and Marcinkiewicz have their idiosyncratic disadvantages. The choice between them will thus always seem unobvious, debatable, and controversial. In such a situation, an effective falsification of results based on one of them by means of an analysis emphasizing the other one is virtually impossible. This leads to a conclusion that we need more reflection and scholarly debate on how to study candidates’ electoral success in complex settings such as multidistrict OLPR systems. Perhaps we ultimately need new, less controversial measures and operationalizations of electoral success in such systems. In the absence of these, students of elections might long be unable to examine unequivocally the soundness of certain contentious concepts, including the “paradox of gender quotas” and related notions.

The second reservation with respect to Jankowski and Marcinkiewicz's model is perhaps more fundamental than the one explicated earlier. While estimating their model of a candidate's electoral success, the authors go to great lengths to control for a candidate's ballot position. They include five dummy variables, indicating the first, second, third, fourth, and last position on a party list, in addition to a candidate's exact number on a list. Meticulous as it certainly is, this analytical strategy can be cause for serious concern and a matter of a methodological debate. This is because the existing literature on the effects of ballot position on a candidate's electoral success in OLPR systems maintains that, more often than not, ballot order is an endogenous variable, and thus its effect on a candidate's electoral success is largely spurious. More precisely, it can be said that “it would not so much be ballot position that explains electoral success but the inverse: the ballot position reflects the potential electoral success” (Lutz Reference Lutz2010, 175). In line with this, empirical studies of ballot position effects in OLPR systems tend to demonstrate that when ballot order is alphabetical or otherwise pseudo-random, then such effects are either absent (Ortega Villodres Reference Ortega Villodres2003) or largely limited to the top (first) position (Faas and Schoen Reference Faas and Schoen2006; Lutz Reference Lutz2010).Footnote ³ While controlling for endogenous covariates almost inevitably leads to biased estimates, in this case, the danger is particularly severe. If gender affects a candidate's potential for electoral success, then both the aforementioned variables will be correlated with ballot position. Therefore, controlling for ballot position in such a manner as is pursued by Jankowski and Marcinkiewicz would almost certainly lead to the attenuation of the effects of gender on a candidate's electoral success. Such a result is indeed arrived at by the authors.

Because of the aforementioned endogeneity concerns, however, one might again be reluctant to interpret the result as effectively falsifying our findings. At the same time, while one might also wonder why Jankowski and Marcinkiewicz do not reflect on the likely endogeneity of candidate ranking, it is also clear that controlling for ballot position in models explaining electoral success (as they do) is, to an extent, defendable. After all, even the overwhelmingly endogenous relationships might contain some genuinely causal components. If such is the case of candidate ranking, then dropping it entirely from the model of candidates’ electoral success would result in omitted variable bias and thereby, again, at least partially invalid findings. Our approach was thus based on a certain “compromise,” emphasizing a rather modest way of controlling for ballot order.Footnote ⁴ Is this “compromise” unquestionably “better” than the analytical strategy that Jankowski and Marcinkiewicz opted for? If one takes seriously into account the extant literature on ballot position effects, then one would certainly be hesitant to answer the foregoing question definitively. In other words, we are again left in the realm of the unobvious, the debatable, and the controversial.

A conclusion here thus necessarily echoes the one drawn earlier with respect to the measures of electoral success: we need more research, more reflection, and more evidence regarding the determinants of electoral success in OLPR systems. In particular, we need to acquire more evidence on the impact of ballot position in such systems. This would ultimately make concepts such as the “paradox of gender quotas” somewhat more falsifiable than they currently seem to be.