Looking for locals under a closed-list proportional representation system: The case of Portugal

Parties’ motivation to include locals depends on th e c aracteristics of the electoral system. This article analyzes where on the lists parties ch oose to position locals under closed-list proportional representation systems. Furthermore, i t also investigates how that choice varies depending on two key district factors, namely distr ict magnitude and whether or not the district is in a peripheral region. To that end, it draws on an original biographical data set of candidates for the Portuguese National Parliament ( 1983, 1995, 2002, 2005 and 2009). The findings show that there are fewer “locals” among t he higher positions on the candidates’ lists and among heads of lists. District magnitude and pe riph ral region effects go in the expected direction, but these effects are small.

Parties' motivation to include locals depends on the characteristics of the electoral system.
While politicians need to signal their personal vote-earning attributes (PVEA) to get elected in open-list proportional representation systems (OLPR), under closed-list proportional representation systems (CLPR) it is the party profile that matters. Here, political parties present ranked lists and voters simply chose one list over another. Therefore, this system offers fewer incentives for political parties to consider PVEA such as localness in the process of candidate selection (Carey and Shugart, 1995); and this is specially the case as district magnitude increases (Shugart, Valdini, and Suominen, 2005;André and Depauw, 2012).
Although, it is more likely to find locals in OLPR systems, it has been shown that even in the most extreme cases of CLPR, such as Israel and the Netherlands, there are significant distinct patterns of geographic representation (Latner and McGann, 2005). This suggests that there might not be many candidates with important local ties within CLPR systems, but there are certainly some. This is so because voters "may be more inclined to vote for a party given knowledge that someone they know from local service is on the list and has a rank that makes election plausible" (Nemoto and Shugart, 2013: 3). Despite some important contributions, many open questions remain on why parties recruit locals in CLPR. This article will address two of those questions: where on the list are locals positioned? And does the probability of finding locals vary significantly across key district characteristics?
The relevance of these questions is two-fold. First, since the voters have no say in the final composition of the lists in CLPR systems, the most important distinction between the candidates is where on the list they stand. Thus, secure positions and especially the top position on the list are crucial for both parties and candidates' aspirations. The head of list is generally well known as she/he is usually the protagonist of the party campaign in the district, and in some cases even at the national level. She/he often has a safe position, particularly in the case of major parties, but more importantly, she/he is the "face" of the party and therefore represents the image the party wants to give itself. She/he personalizes the party's campaign and, at least in theory, her/his characteristics are considered the best to enhance the party's performance. If the "value" of the candidate somehow translates to where she/he is positioned on the list, then it is likely that localness will vary along the positions on the list. The expectation that candidate order is relevant for parties and voters is mostly explored in open list systems (Chen et al., 2014), but we argue that it also matters in CLPR and in the 1 The authors would like to thank the reviewer for the careful reading of the article and for the precise and insightful comments.

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
3 Portuguese case in particular, where studies on leaders effects (Lobo, 2006) and modes of campaign (Lisi and Santana-Pereira, 2014), suggest that citizens do value candidates' characteristics.
Second, district characteristics may influence the probability of finding locals. Beyond district magnitude, considering whether or not the district is in a peripheral region might prove meaningful for capturing the specificity of recruitment in areas where political attitudes and demands are tendentiously more localized.
The article draws on an original data set of very comprehensive biographical data on candidates for the Portuguese National Parliament. The data set comprises all winners and many losing candidates 2 of the most relevant political parties who ran for five legislative elections between 1983 and 2009. The Portuguese case offers the perfect conditions to analyze these research questions, as it features a steady CLPR system with great variance at the district level, namely concerning district magnitudes and with two out of 22 districts that are clearly peripheral regions (Azores and Madeira).
To answer these questions, we have organized this article as follows: the next section presents the state of the art about localness and candidate selection in CLPR systems and introduces our working hypotheses; Sections 3 introduces the key institutional features of Portugal and resumes what is already known about the way parties chose their candidates; Section 4 details the structure of the data and shows descriptive results on localness and main explanatory variables; Section 5 presents the results of the main analysis; and finally Section 6 summarizes the main conclusions of the study.

Localness within CLPR systems: working hypotheses
One of the shortcuts citizens employ to elect legislators is that of local ties, as these are cues to politicians' understanding of local needs. While party platforms convey information to voters about the policy goals of parties, candidates' attributes may signal more parochial and local considerations (Shugart, Valdini, and Suominen, 2005: 440). Furthermore, diverse indicators of localness might fulfill slightly different goals: '(…) local birth can be viewed as a way of communicating to voters that «I know what you want», [electoral] experience is a way of reminding voters that «I know how to get it»' (Shugart, Valdini, and Suominen, 2005: 441). Tavits adds that it is more credible for a candidate with local roots to claim to be a local servant in national decision-making than for one without local roots (2010: 217). As a 2 See footnote 15 for further details on losing candidates.
consequence, candidates with strong local ties have better chances of being elected, at least in systems that allow for intraparty competition such as in Ireland and Estonia, and majoritarian systems like Canada's (Marsh, 1987;Blais et al. 2003;Tavits, 2010) 3 .
What about closed-list systems? Considering that voters are forced to vote for a whole list with a pre-determined order, what could justify a party's strategy to nominate locals? The literature offers some interesting answers to this question. A first approach argues that the simple fact that there are sub-national electoral units in most countries creates a basis for regional representation (Latner and McGann, 2005: 712). Another view underlines the electoral advantage usually associated with balanced lists from all perspectives, but especially from the regional perspective (Latner and McGann, 2005: 713;Nemoto and Shugart, 2013: 3) 4 . Finally, party organizations from different regions are likely to apply pressure for having "their" candidates in good positions on the lists (Latner and McGann, 2005: 713). This should especially be true if the recruitment process is a decentralized one. Comprehensibly, parties with decentralized organization are expected to nominate more local candidates than centralized parties (Hazan and Rahat, 2010).
However, altogether, the reasons for including locals as candidates are by far more salient in open than in closed-list systems. In fact, although there is some evidence of the presence of locals in CLPR lists (Latner and McGann, 2005 for Israel;Shugart, Valdini, and Suominen, 2005 for other countries), there is a broad consensus around the idea that personal vote-earning attributes such as localness are not very valuable in closed-list proportional representation systems (Carey and Shugart, 1995). This is the case not only because being local offers limited electoral utility, but also because locals might imply other costs for the parties that in the end do not pay for their inclusion. Using a diverse, though small, set of countries, Tavits (2009) argues that MPs who have local-level political experience are more likely to behave independently, and therefore vote against the party line in parliament (see also Tavits, 2010).
Hence, our first hypothesis poses: if localness is indeed not a relevant asset in CLPR systems, the more secure the position on the list, the lower the probability of finding locals (H1a). Furthermore, since heads of lists occupy the most secure positions of all, we expect that the lowest probability of finding locals is reached within the first positions on the lists (H1b).

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
Within the literature on the impact of electoral rules on personal vote -defined as part of the legislator's vote that is based on his or her individual characteristics or record (Cain, Ferejohn, andFiorina, 1987 through Shugart, Valdini, andSuominen, 2005: 437)the ballot structure and the district magnitude have privileged positions. Long ago, Carey and Shugart (1995) argued that the kind of ballot and district magnitude interact to affect personal vote. One decade later, the argument was empirically corroborated. When lists are closed, legislators' PVEA (operationalized as local birthplace or lower-level electoral experience) are of decreasing usefulness to voters as magnitude (and hence the number of candidates on a list) increases. Conversely, in open lists the PVEA usefulness is greater as magnitude increases (Shugart, Valdini, and Suominen, 2005). Although they use a different measure of localness as a dependent variable, namely the time legislators spend nurturing their districts, André and Depauw (2012)  That happens mainly for two reasons. The first one is that since there are fewer candidates, they are arguably easier to get to know and therefore the people pay more attention to who they are. The second reason is that high magnitude districts coincide with more urban areas, which are usually more impersonal, not only because of the bigger dimensions but also because their inhabitants tend to have lighter relationships with these districts -many were born somewhere else. If this special attention towards PVEA (including localness) existed in small magnitude districts, we would expect to find locals in more secure positions in those districts. Therefore, our second hypothesis poses: the higher the district magnitude, the lower the probability of finding locals on candidates' lists (H2).
Among low magnitude districts, those in peripheral regions enjoy some particular characteristics. Latner and McGann (2005) have shown that even in cases of extreme proportionality -such as Israel and the Netherlands -there is some representation of geographic interests, especially in peripheral regions. This happens because those regions usually have strong regional identities and political interests, so parties tend to consider being local as an asset to capture votes. In fact, in those districts voters arguably pay more attention to the candidates' characteristics that prove they belong (and are committed) to the district. In the case of Portugal, Azores and Madeira fit well within the category of peripheral regions.
Their geographical remoteness, their high level of autonomy, insularity, and isolation, their low dimension, and the fact that they are considered as ultra-peripheral by the European Union leave little doubt. Moreover, MPs from Azores and Madeira show particular interest in representing the local needs of their electors (Leston-Bandeira, 2004: 143-144). Hence, our third hypothesis argues that there is a higher probability of finding locals on the candidates' lists of districts in peripheral regions (H3).

Portuguese Context
Portugal is a unicameral semi-presidential regime with a proportional electoral system using a closed-list ballot, adopted from the onset of democratic transition. The unicameral parliament, named Assembly of the Republic, is currently composed of 230 seats, which are allocated through the application of the D'Hondt method in 22 plurinominal districts. The average district magnitude (11) is relatively high when compared to other European democracies such as Spain and France (Jalali, 2003: 548), yet there is great variance across these units: the largest district (Lisbon) currently elects 47 seats and the smallest elects as few as two seats.
Even though the issue of electoral reform is recurrently put into the public agenda, the only changes made so far have to do with the sizes of both the parliament 5 and the electoral districts (and these are rather minimal, i.e. the allocation of one or two seats may change in between elections). All other rules have remained unchanged.
Thus, it is a stable institutional framework that has been the basis of electoral Studies about parliamentary recruitment, and more specifically candidate selection in Portugal are rather recent and have mainly focused on the type of party recruitment (decentralized vs. centralized) and on the outcome of political recruitment. To specify, they have focused on depicting the political and socio-demographic profiles of the candidates elected for parliament (Cruz, 1988;Freire, 1998;2001;2002;Freire, Matos, and Sousa, 2001;Freire and Pinto, 2003;Freire, Viegas, and Seiceira, 2009). Nevertheless, little is known about the way PVEA, namely localness, are weighted within parties' lists.

M A N U S C R I P T
A C C E P T E D ACCEPTED MANUSCRIPT Teixeira's (2009) 7 study about the candidates' for parliament between 1987 and 2002 is one of the few that tackles the localness issue, since she explores, albeit superficially, the relationship between district magnitude and the candidates' local ties. The results are as interesting as they are surprising, insofar as they show that localness matters: (i) 94% of the candidates are insiders, meaning that they are either born and/or reside in the same district in which they are running for elections 8 , (ii) there is a positive correlation between district magnitude and localness, with insiders being more frequent in large and medium size districts, and (iii) there is a correlation between being local and having higher positions on lists, because insiders are more common in eligible positions than in non-eligible positions.
These findings portray the Portuguese case as puzzling, given that it contradicts two solid assumptions, namely that CLPR systems offer few incentives for political parties to care about the PVEA (namely localness) of their candidates and that there are fewer locals at high district magnitudes. These results are even more significant if we take into account that Shugart, Valdini, and Suominen (2005: 444) demonstrated the opposite using biographical data of candidates elected to the Portuguese parliament in 1995: the probability of being native to the district decreased significantly as district magnitude increased 9 . Yet, it should be noted that these studies covered different time periods and treated localness differently.
Whereas Teixeira (2009) covers five elections and combines being native and resident under the same category (which the author calls insiders), Shugart, Valdini, and Suominen (2005) observed one election and estimated the probability of being native and of having lower-level electoral experience individually. The present article adds new empirical findings and gives a valuable contribution to the literature on this topic by helping to disentangle this puzzle.
In our view, tackling this puzzle requires two important decisions from the outset of the analysis. The first relates to how localness is operationalized. Like Shugart, Valdini, and Suominen (2005), we believe that each component of localness signals different things to voters and thus matter differently for parties. In this sense, clustering indicators of localness together might overshadow an important source of variance. Otherwise, treating them individually might prove more meaningful for understanding parties' rationales for selecting candidates according to different types of local ties. Furthermore, our data set is the first one 7 The study relies on biographical data collected at National Electoral Commission (CNE) for the country's major parties and alliances. 8 This result is even more surprising as all but one of the main Portuguese parties has a centralized political recruitment process. However, the measure of localness used by the author is very broad: "born and/or resident in district." In our data set, selecting only the most important parties, the respective value is 81%. 9 They found no significant result for lower-level electoral experience in the case of Portugal (Shugart, Valdini, and Suominen, 2005: 444).
on Portugal that includes the three main indicators of localness: being native, being resident, and having experience in local politics.
The second relates to the data collection process. We have used a bigger and more varied number of sources than the existing data sets on candidates to Portuguese parliament; therefore, our data are more complete. Besides drawing upon the electoral forms deposited at the National Electoral Commission (CNE) 10 , and the public biographies of MPs (either published and/or available online through the parliament website) 11 as all other previous studies have done, we have also collected information available online on each person (through Google). Moreover, for the lower-level electoral experience 12 we did systematic matching between the names in our data set and a list of those elected for local positions. Shugart, Valdini, and Suominen (2005: 439) wrote: 'Carey and Shugart's (1995) logic takes the perspective of the politician, cultivating (or not) a personal vote. We extend this logic further by tying it to the demands of voters.' In the present article, we take yet another perspective by focusing on the parties' strategies -where parties choose to place locals and whether this strategy varies significantly according to district characteristics. The rational for the option to focus on parties is connected to the fact that CLPR systems are clearly partybased systems. So, although it can be argued that even in these systems, politicians might choose whether or not to cultivate their personal vote and that voters might decide which party to vote for because of candidate attributes, what really matters is the position each candidate occupies on the list. That is something that is largely decided by the party leadership.

Data structure, measurement of variables, and descriptive results
To research the questions raised in this study, we use an original data set of candidates for the Assembly of the Republic that we put together 13 . The data set covers five parliamentary elections in 1983, 1995, 2002, 2005, and 2009 14 and includes biographical background data on the candidates of the six most relevant Portuguese political parties. We collected biographical data on all winning candidates. Data were also collected for the most significant 10 Which are not public, but can be consulted upon request. 11 The Portuguese parliament does not impose any form on the MPs' CVs. They are free to post whatever information they wish on the biographical pages. Those biographies are eventually gathered into books. 12 As well as for all other political positions -although they are not that relevant for this paper. 13 Reference of data source is omitted for blind peer review. 14 The choice to collect data on these particular elections was based on a comparability criteria with the other cases included in the project.
losers from the major parties (PS and PSD) 15 and for the first losing candidates from secondary parties (CDS-PP, PCP, PEV, and BE). The basic observation unit for this data set is candidate election, and this gives us 1759 candidate election observations in total.
Before proceeding to the main analysis, let us introduce the variables used in this study and briefly show how they are portrayed by our data.

Dependent and Independent Variables
We follow the research design proposed by Shugart, Valdini, and Suominen (2005) and therefore, we define the probability that a candidate has local ties (localness) as our dependent variable (DV) 16 . As mentioned before, to operationalize localness, we use three variables.
Two of these have often been used in the literature, namely whether or not the candidate was born in the district where she/he runs (native) and whether or not the candidate has lowerlevel electoral experience (local politics) 17 . The third one, which states whether the candidate lives in the district (resident), is less common in the literature. These three are dummy variables.
As mentioned before, we are particularly interested in the probability of finding locals according to the positions the candidates occupy on the lists (i), among heads of lists (ii), depending on the district magnitude (iii) and in peripheral regions (iv). These comprise our four independent variables. Positions on the lists were operationalized considering each candidate's relative rank, which was calculated by subtracting the position each candidate occupies on the list from the number of MPs her/his party elected in that district. For example, if Candidate A occupies position number 4 in the district of Lisbon and her party elected 10 MPs in that district, then her relative rank is 6 (10-4). By contrast, Candidate B was placed in the first position of that list, so her relative rank is 9 (10-1). Therefore, the higher the relative rank, the more secure the position. Candidates with negative numbers as relative ranks did not get elected. For example, Candidate C occupied position number 30 on the same list, so her relative rank is -20 (10-30), i.e. she was 20 positions away from succeeding into parliament 18 . This is a discrete numerical variable that goes into our data set with values ranging from -6 to 23. Concerning heads of lists, a dichotomous variable was constructed, where 1 corresponds to the first position on the list, and zero to all other positions. District magnitude corresponds to the number of MPs elected in each electoral district. Although the range of district magnitude currently runs from two to 47 MPs (Lisbon), it used to go up to 56 (in 1983).
Therefore, this discrete numerical variable runs from two to 56 MPs in our data. Finally, the variable peripheral region consists of a dummy variable for the districts of the autonomous regions of Azores and Madeira. Table 1 displays the global and yearly relative rank average of those who are local and of those who are not. Generally speaking, we observe significant (though not very substantive) differences between locals and non-locals in terms of the average position on the lists. However, there are differences contingent to the type of localness. Those who are native to the district and who have experience in local politics have lower relative ranks (i.e. have less secure positions) than those who do not share those characteristics. On the contrary, candidates resident in the district have better positions on the lists than non-residents. In fact, residents feature on average a relative rank of 2.2 whereas non-residents have 1.6. This pattern is present when the election years are considered separately, although the differences are only rarely significant. Overall, it is a consistent but soft pattern.  candidates. This time the data are shown as percentages. There are considerable fewer locals among heads of list vis-à-vis all other candidates, either generally -native (39% against 57%), resident (57% against 86%), and local politics (35% against 56%) -or when we look at yearly percentages. This suggests that political parties are privileging other criteria (such as party loyalty, political career, and public notoriety) rather than local ties when selecting the heads of lists. Furthermore, we do not find a great variety across elections, although two light evolutionary tendencies might be observed. The first one is that, compared to the 1980s, heads of lists today are less native than they used to be. The second tendency concerns local politics, which increases both for heads of lists and for all other candidates over time, and particularly among the former. This might be related to the consolidation of the Portuguese political system, i.e. it is very likely that in the beginning of the democratic period, few candidates with local political experience were nominated simply because there were fewer potential candidates with that attribute. Finally, place of residence emerges as the most relevant indicator of localness, since its percentage is systematically higher across elections (consistent with the results for relative rank).
Altogether, our data suggest that Portugal features high percentages of candidates with local ties (thus confirming Teixeira, 2009). Shifting to the district characteristics in which locals and non-locals run, Table 3 presents average district magnitudes by type of localness and year. The results highlight two things. First, being resident is the only type of localness for which we find systematic and significant differences in terms of the average district magnitude, with resident candidates running in larger circles (22 seats) than non-residents (14 seats). The same pattern can be observed for every election year included on the table. Second, the remaining two indicators of localness have no apparent and significant relationship to district magnitude (the partial exception is local politics in 2005).  Table 4 displays the percentage of locals among the candidates who ran in peripheral districts compared to those who ran in non-peripheral ones. Here again we conclude that localness is not homogeneous. We find significantly higher percentages of natives (82.6%) and residents (92.3%) in parties' lists of peripheral regions than in nonperipheral regions. However, we find the opposite result with regards to local politics: nonperipheral regions feature more candidates with experience in local politics than peripheral ones. Note: Percentages were calculated for the total of candidates in our data set who ran in peripheral districts (and in non-peripheral districts) within each year. For example, in 1983, 66.7% of candidates who ran in peripheral districts were born in that district. Numbers in bold indicate significant differences between the percentages of locals between peripheral and non-peripheral regions as a result of a chi-square test.

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
Besides the four main independent variables, we have employed several controls, which provide fairly robust models. First, we estimated models for each type of localness in which we controlled for the effect of other types of localness. So for instance, if being native is the DV, then we control for whether the candidate is resident and has experience in local politics in the district where she/he is running for election. Second, some socio-economic variables were also included, namely year of birth (discrete numerical), being a women (dummy variable 1 = woman), and level of education (a categorical variable with 5 categories). Of utmost importance is the inclusion of the variable "incumbent" (dummy variable 1 = incumbent 19 ), which accounts for the higher probability that incumbent candidates have of being nominated to higher positions on lists (Schwindt-Bayer, 2005).
Finally, a last control variable was added for major parties (dummy variable 1 = major party) because the political strength of a party is likely to play a crucial role on the allocation of candidates to the lists (Nemoto and Shugart, 2013: 4). In particular, while heads of lists of major parties have guaranteed places as MPs in almost all districts, the same is far from true concerning secondary parties, which have small chances in low magnitude districts. 19 A candidate was considered incumbent when she/he was serving as a member of parliament when the candidates' lists were built.

M A N U S C R I P T
A C C E P T E D ACCEPTED MANUSCRIPT 14

Main analysis
As mentioned before, following Shugart, Valdini, and Suominen (2005), we use the probability that a candidate has local ties as our dependent variable, which originates three dependent variables: the probability (Pr) that a candidate is native to the district -Pr (native), the probability that a candidate lives in a district -Pr (resident), and the probability that a candidate has lower-level political experience in a district -Pr (local politics). Having this as a baseline, four models were estimated: one including the variable that identifies the relative position occupied by the candidate on the list, relative rank (Model 1); another indicating whether or not the candidate is a head of list (Model 2); a third one having district magnitude as the main independent variable (Model 3); and a final one where the most relevant factor is whether or not the region is peripheral (Model 4). We estimate our models using probabilistic regression (probit), with fixed effects for election years and with errors clustered at the district level, due to the multilevel nature of the data. Tables A1, A2, A3, A4 in the Appendix present the results for the four models, respectively.
Looking at the overall health of each model (namely at the Pseudo R2), we conclude that our models succeed more at predicting the probability of finding district residents than at predicting the probability of having a native candidate or a candidate with local political experience. The models confirm that localness is not a homogenous concept leading to mixed results for all hypotheses. We now proceed to show the most relevant predicted probabilities 20 in order to get a clearer idea of the results we get for the most important independent variable in each model. When reporting the results, we consider that the probability of finding locals is significantly affected by a given independent variable when the confidence intervals associated with the predicted probabilities (for each value of the independent variable) do not overlap. 20 Using Clarify, in Stata.

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
15

Figure 1 -Predicted probabilities of being local (native, resident, and local politics) depending on the candidates' relative rank (Model 1).
Note: The values in axis x correspond to the original distribution of the variable in the data set.
According to our first hypothesis (H1a), the more secure the position on the list, the lower the probability of finding locals. Looking at Figure 1, we observe mixed results. On the one hand, we confirm that being native and having experience in local politics matter little when party leaders are selecting candidates to good positions on lists, in line with Carey and Shugart (1995). The result concerning local politics comes as a surprise since it contradicts the common perception in Portugal that local political experience is relevant in parliamentarians' careers (Freire, 2002). According to our data, that experience does not assure candidates a higher chance of getting into parliament. On the other hand, being a resident seems to be a valuable asset. In the third section "Portuguese Context", we mentioned that Teixeira (2009) concluded that locals are more common in eligible than in non-eligible positions. The fact that this author aggregated being native and living in a district in a single variable might have led her results to be driven by residency. In fact, when both indicators of localness are kept separate (Figure 1), they portray contradictory results: while residency is an added value, being native is quite irrelevant.

Figure 2 -Predicted probabilities of being local (native, resident, and local politics) for heads of lists and for all other candidates (Model 2).
Notes: The values selected for the x-axis correspond to the current distribution of the number of MPs by district in Portugal.
According to hypothesis H1b, the lowest probability of finding locals is reached when heads of lists are considered, since they occupy the most secure positions of all. The results presented in Figure 2 give partial support to this hypothesis 21 . The probability of having a local occupying the first position on a list is usually lower than the probability of finding locals among all other candidates for two kinds of local ties, namely residency and having local political experience. However, the latter result is not significant and being native is not a differentiating criterion separating heads of lists from all other candidates. We argue that the opposing results we get for being native and resident when secure positions or heads of lists are considered are related to the way the political recruitment takes place in Portugal. This issue is given full attention below when we discuss Table 5. 21 To control the robustness of our findings, we estimated Model 1 and 2 again (i.e. H1a and H1b), having relative rank and head of list as dependent variables and all local variables as independent variables plus controls. Our findings hold overall. The only difference is that residency is not significant in the first model of the alternative model. The reason why we opt to present our models with locals as a dependent variable is to allow greater comparison with the findings of Shugart, Valdini, and Suominen (2005).

Figure 3 -Predicted probabilities of being local (native, resident, and local politics) depending on the district magnitude (Model 3).
Figure 3 presents predicted probabilities for each of the dependent variables according to district magnitude, whose effects we test under H2. Our expectation for this hypothesis is that the higher the district magnitude, the lower the probability of finding locals. Looking at the figure, it becomes clear that the district's dimension is not a very differentiating criterion for finding locals, and once again we get mixed results. Although the curves for being born in a district and having experience in local politics run in the expected direction -as in Shugart, Valdini, and Suominen's cross-national analysis (2005) -in our data, the differences between low and high district magnitude are not as striking, particularly for local politics 22 . This lies in the fact that, in Portugal, the first names on each list are often chosen by the national party leaders (Freire and Teixeira, 2011). Whereas this leaves room for district party branches to have a say in who receives many eligible positions in high magnitude districts, in much smaller magnitude districts (say 2, 3, or 4 positions) all or almost all eligible places are centrally chosen. Since party leaders tend to select non-locals, this contributes to decreased differences between district magnitudes. 22 Looking at Figure 3 we realize that the difference in the probability of being native is only significant when the highest and lowest positions on the list are considered.

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
Furthermore, residency behaves conversely, i.e. the probability of finding residents increases as the district magnitude grows, in line with Teixeira (2009). This might happen simply because the higher magnitude districts correspond to the two biggest cities (Greater Lisbon and Porto) where more potential candidates live. That is very likely the case since, according to our data, those are clearly the cities where the highest percentages of candidates live, with 33% and 13% respectively.

Figure 4 -Predicted probabilities of being local (native, resident, and local politics) for peripheral and other regions (Model 4).
The expectation that parties nominate more locals for peripheral regions than for nonperipheral ones (H3) is confirmed for Pr (native) and Pr (resident), although only the former has significant differences in predicted probabilities (see Figure 4). The probability of being native -and to a certain extent resident -is higher in peripheral regions. However, when we look at Pr (local politics) we realize that it is actually higher in non-peripheral regions.
This implies that being local in a peripheral region mainly means being native, partly means living there, but not at all means having served in local politics. Since Azores and Madeira have the status of autonomous regions, they also have their own political organs, namely regional parliaments and governments, which are elected in specific elections. Hence, it could M A N U S C R I P T

A C C E P T E D ACCEPTED MANUSCRIPT
19 be the case that the candidates of these districts to National Parliament did not have lowerlevel electoral experience, but did have regional-level electoral experience. In order to test for that possibility, we added information on whether each candidate was ever elected to the Legislative Assembly of the Autonomous Region or nominated to the regional government to the variable local politics (dummy variable 1 = ever elected/nominated). The results are shown in the last graph of Figure 4 and in the fourth column of Table A4. Indeed, when both regional and local political experiences are considered, the difference between non-peripheral and peripheral regions substantially decreases and stops being significant. Note: "Yes" identifies that the predicted probability for a given indicator of localness follows the hypothesis ("No" otherwise). Bold is used to signal significant predicted probabilities. † This result is probably due to the fact that the higher magnitude districts correspond to the two biggest cities and these are the cities in which the highest percentages of candidates live.
As summarized in Table 5, the results vary substantially depending on how localness is measured, and none of our indicators of localness confirms all hypotheses. Being resident is the measure of localness that challenges more hypotheses, whereas being native and having local political experience fit our expectations better. Furthermore, since the latter two help sustain the more classical hypotheses (H1a and H2), they seem to better correspond to the established knowledge on the localness of candidates to National Parliaments which has been very much informed by Carey and Shugart (1995). In other words, residency's inclusion as an indicator of localness covers a dimension of candidates' localness that has been overlooked by the literature.
Our data show that there are different ways of being local, reinforcing the idea that each indicator of localness passes on a diverse message to voters and is used differently by the parties. Plus, not all gatekeepers perceive localness equally, even within the same party. More precisely, our data suggest that, whereas national party leaders prioritize native candidates, district leaders prefer residents, as explained below.

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
As mentioned before, political recruitment in Portugal is a fairly centralized process (Freire and Teixeira, 2011)  Whereas heads of lists are nominated by the national leaders alone, secure positions are nominated both by national leaders and district party leaders, which has implications in these candidates' localness. Since H1a is confirmed for being native and having lower-level experience but not for being resident, we argue that district party leaders prefer to choose residents. Many reasons might be behind this preference. Resident candidates supposedly know their districts and the people they are about to represent, have proven to be loyal to the party district, and their nominations are easy to justify to constituents. Plus, a resident candidate has possibly acquired "local capital" through the time she/he has spent in the district either studying or working and therefore her/his nomination might increase the odds of the party getting a good electoral result. Though this might be also true for candidates with local political experience, our data suggest that the pool of potential candidates to National Parliament is greater than the group of local politicians, implying that only a few local politicians are nominated for secure positions to National Parliament.
The latter sentence applies particularly well to peripheral regions, as illustrated by the negative result we get for local political experience in our final hypothesis (H3). In fact, M A N U S C R I P T

A C C E P T E D ACCEPTED MANUSCRIPT
parties do not seem to value candidates' local political experience in these districts as much as expected. Otherwise, what seems to count more for candidates' recruitment -although it does not count very much -is their career at the regional level (parliament and government).
A last note on the second hypothesis (H2): we have intentionally left it out of this summary, since the only outlier result (residency) for this hypothesis (as previously argued) is due to the circumstantial reason that the higher magnitude districts correspond to the two biggest cities where more candidates live.

Conclusions
Focusing on Portugal's CLPR system and on parties' strategies for selecting locals, this article  (Carey and Shugart, 1995). The fact that localness is lower among the higher positions on the candidates' lists and among heads of lists demonstrates that candidates' local ties are not valuable assets for Portuguese political parties.
The only clear exception to this tendency is that resident candidates are easier to find as we go up the positions on the lists. We will come back to this exception below.
In the second question, we examined the probability of finding locals by district characteristics. The analysis developed here suggests that district magnitude has a rather minor role in explaining locals' presence within party lists. The only exceptions are native candidates, who have lower expected chances of being selected as district magnitude increases. Otherwise, having experience in local politics is irrelevant and being resident works in the opposite direction, i.e. we find more resident candidates at higher district magnitudes.
Our results show a less clear picture of the effect of district magnitude on candidates' localness in CLPR systems than previous studies have shown (Shugart, Valdini, and Suominen 2005). We believe that some intraparty variables -in particular characteristics of the recruitment process -intervene here, corroborating the idea that party organization plays a role in explaining candidate selection (Hazan and Rahat, 2010). As mentioned before, due to M A N U S C R I P T A C C E P T E D Still concerning the second question, the specificity of peripheral regions explains the significantly higher number of native candidates in those regions, reversing the results stated before for being native and district magnitude. However, the centrality of the recruitment process is strong enough to avoid the same effect taking place for the other two indicators of localness.
Altogether, our results demonstrate that localness is not homogenous and that while some indicators of localness support the well-established findings in the literature on this matter (in particular being native, but also having local political experience), others (namely being resident) challenge them. We have previously mentioned the probable implications for voters of being native and having local political experience, namely, that the candidates are aware of the constituency's problems (native) and that they can solve them (local politics) (Shugart, Valdini, and Suominen 2005: 441        Note 1: Standard errors clustered at the district level in parentheses. Election year fixed effects are not shown. †p<0.10; *p<0.05; **<0.01; ***p<0.001. Note 2: In the fourth column, Pr (regional & local politics), the variable local politics also includes regional and governmental political experience.

M A N U S C R I P T
A C C E P T E D