The labyrinth of ethnicity

Before analyzing whether there is a specific policy supply for indigenous legislative candidates, it is necessary to clarify how this type of identities is conceived in the specific context of the country so that we can determine how to capture and measure them. This invites us to review the approaches that have been developed in this regard in history, anthropology, sociology and demography, with tools and concepts that are still under discussion.

What does it mean to be“indigenous”in Mexico?

There is a complex debate in social sciences on how to define ethnic identities. Ethnic elites always refer to supposedly objective criteria to highlight the characteristics that distinguish and separate them from other human groups. However, these attributes are frequently backed up with subjective differentiations that change according to the contexts in which they are stated. From this perspective, ethnic identities are contingent and situational, relational and inter-subjective socio-political constructions. Therefore, the representation of the socio-cultural features that are used to re-produce ethnic boundaries is much more important than the attributes themselves (^{Barth, 1969}).

In Mexico, autochthonous peoples occupy a peculiar place in the symbolic construction of the Nation, since they are part of the constitutive myth that separated the Mexican from the Creole identity of the Spanish Conquerors. It is an identity highly valued in museums and collective memory, which serves as a central referent of otherness in the construction of miscegenation and of the “raza cósmica” (cosmic race), within a post-revolutionary project of assimilationist integration that also tactically associated it with barbarism and cultural backwardness. The word “indio” (indian) is used pejoratively to stigmatize inappropriate behavior, with a strong component of classicism that is not always or necessarily racial. The “indígena” (indigenous) concept, on the other hand, is used in a neutral way to designate the native populations, although it is still loaded with ambiguous and discriminatory connotations too. Hence the need to question some erroneous ideas strongly rooted in collective imaginations.

When seeking empirical approaches to these populations, the multi-dimensionality of the concept comes to the fore, in contrast to the Manichean character of the predominant social representations. From a historical perspective, the relevant criterion for classification is membership in some pre-Hispanic community. However, it is not easy to establish which groups in fact meet this requirement, since the processes of conquest, colonization and independence were accompanied by sociodemographic transformations that profoundly reshaped the country. Historians thus acknowledge the existence of several thousand “peoples” (formerly called “República de indios” or “Republics of Indians”) whose origin dates back to the Conquest or, in many cases, colonial times (^{Warman, 2003}).

From a cultural perspective, usually the distinctive feature is speaking an indigenous language. Learning a language —and passing it on to your children from early childhood— is a strategic decision that involves years of structured interaction and only makes sense when that skill represents a need or an effective advantage. As Figure 1 illustrates, the number of Mexicans over the age of five who declare that they speak an indigenous language has increased considerably in the last 120 years, going from 2 030 714 in the 1895 Census to 7 386 791 people in the Inter-Census Survey of 2015. During the same period, its proportion decreased in relative terms, going from 16 per cent to less than 8 per cent from 1990, stabilizing around 6.5 per cent between 2010 and 2015. Even more noteworthy is the continuous reduction of the population that does not speak Spanish: it went from 1 794 306 to 910 053 people between 1895 and 2015 (that is, from 15.4 to 0.8 per cent of the total).

Source: Own elaboration based on general population censuses (INEGI, 1895, 1910, 1921, 1930, 1940, 1950, 1960,v1971, 1981, 2001 y 2011) and the Intercensal Survey 2015 (^{INEGI, 2015}).

FIGURE 1 Quantitative evolution of the speakers of indigenous languages (1895-2015) 

Upon the risk of underestimating the indigenous population, alternative estimates have been developed. Among these, the “Indigenous population in indigenous households” (“Población indígena en hogares indígenas”) stands out. An indigenous home is defined as “one where the head of the household, his or her spouse or one of the ancestors (mother or father, stepmother or stepfather, grandmother or grandfather, great-grandmother or great-grandfather, great-great-grandmother or great-great-grandfather, mother-in-law or father-in-law) declared to speak an indigenous language”. In 2015, a total of 11 938 749 people, namely 10 per cent of the Mexican population, were indigenous according to this definition (^{CDI, 2017}).

More recently, new survey questions have been experimented with to estimate the so-called “self-affiliation”. A first exercise was carried out with a sub-sample of the INEGI Census in 2000, which captured 5.3 million people who “considered themselves indigenous”. This was replicated in the 2010 Census and in the 2015 Inter-Census Survey, with a substantive modification of the phrasing that expanded its meaning by asking if “According to your culture, (name), do you consider yourself indigenous?” (INEGI, 2016). The underlying problems with these methodologies are reflected in the evolution of the resulting estimates that stand in acute contrast to the evolution of the percentage of speakers of indigenous languages. As can be seen in Figure 2, the proportion of people who self-describe as indigenous increased exponentially, going from 6.1 to 14.9 per cent and to 21.5 per cent between 2000, 2010 and 2015.

Source: Own elaboration based on general population censuses (INEGI, 1895, 1910, 1921, 1930, 1940, 1950, 1960, 1971, 1981, 2001 y 2011) and the Intercensal Survey 2015 (^{INEGI, 2015}).

FIGURE 2 The growing gap between linguistic and self-description criteria (2000-2015) 

To these official estimates we must add the Barómetro de las Américas or Americas Barometer surveys. When asked “Do you consider yourself a white, mestizo, indigenous, black, mulatto, or some other?”, between 6.9 per cent (in 2012), 11 per cent (in 2014) and 9.5 per cent (in 2017) of the respondents answered that they considered themselves “indigenous people”. However, when asked “According to your culture, do you consider yourself indigenous?”, between 43.3 per cent (in 2014) and 47.2 per cent (in 2017) of the same respondents answered affirmatively (^{LAPOP, 2012-2017}).

Rather than being a byproduct of demographic revolution, this growing gap is the result of different understandings and it reveals a gradual transformation regarding the social meaning of the concept “indigenous culture”, which is not reflected in a concomitant expansion of people who decide to transmit an indigenous language to their family members. In this paper, we favor the linguistic criterion, which underestimates the effective number of indigenous people but is more stable and less subjective, in addition to being quantifiable at the finest levels of electoral geography (which is not yet possible for self-affiliation).

Geography and a minimal sociology of indigenous “peoples” and “communities”

The 2010 Census registered 64 ethno-linguistic groups that were concentrated in 803 municipalities, 4 394 electoral sections and 28 338 localities with more than 30 per cent of indigenous language speakers (^{INEGI, 2011}). Five years later, the 2015 Intercensal Survey added six more regional variants, which makes the great linguistic heterogeneity of Mexico worth noting. As Table 1 shows, only 16 groups have more than 100000 inhabitants and only six have a population equivalent to or greater than the average size of a single-member district. In contrast, 36 groups have less than 10 000 inhabitants, 22 have less than 1 000, and five do not even reach 100 inhabitants. Thus, while the Nahuatl people are enough to constitute a small State, other indigenous populations barely reach the necessary magnitude to preserve community autonomy.

TABLE 1  Demographic heterogeneity of ethno-linguistic groups in Mexico 

Indigenous language in Mexico (speakers 3 years and over to 2015)
Indigenous language	Total	Indigenous language	Total	Indigenous language	Total
Náhuatl	1 725 620	Tepehuano del sur	36 543	Other American languages	1 126
Maya	859 607	Cora	28 718	Lacandón	998
Tseltal	556 720	Chontal de Tabasco	27 666	Seri	754
Mixteco	517 665	Triqui	25 674	Pima	743
Tsotsil	487 898	Yaqui	20 340	K’iche’	730
Zapoteco	479 474	Huave	18 539	Chocholteco	729
Otomí	307 928	Popoloca	18 206	Jakalteko	527
Totonaco	267 635	Cuicateco	13 318	Kumiai	486
Chol (Ch´ol)	251 809	Pame	12 232	Texistepequeño	455
Mazateco	239 078	Mam	11 387	Cucapá	278
Huasteco	173 765	Tepehua	10 427	Paipai	216
Mazahua	147 088	Tepehuano del norte	9 568	Kiliwa	194
Tarasco	141 177	Q’anjob’al	8 421	Unspecified Tepehuano	170
Chinanteco	138 741	Unspecified Popoluca	6 122	Ixcateco	148
Tlapaneco	134 148	Chontal de Oaxaca	5 064	Qato’k	134
Mixe	133 632	Sayulteco	4 117	Kickapoo	124
Unspecified indigenous language	101 187	Chuj	2 890	Pápago	112
Tarahumara	73 856	Akateko	2 837	Ixil	103
Zoque	68 157	Chichimeco jonaz	2 134	Oluteco	90
Amuzgo	57 589	Guarijío	2 088	Teko	81
Tojolabal	55 442	Matlatzinca	1 568	Kaqchikel	61
Huichol	52 483	Tlahuica	1 568	Ayapaneco	24
Chatino	51 612	Q’eqchi’	1 324	Aguacateco (Awakateko)	17
Mayo	42 601	Unspecified Chontal	1 135
Popoluca de la sierra	37 707

Source: Own elaboration based on the Intercensal Survey 2015 (^{INEGI, 2015}).

This prompts the question about the most suitable territorial level to study the political behavior of indigenous populations. The 28 000 localities include thousands of scattered hamlets where only a few isolated families reside, making them too small to be considered as culturally autonomous indigenous “peoples”. At the same time, the classic anthropological approach of studying entire municipalities to capture traditional fiefdom systems and community networks of mutual support, which made a lot of sense from the 1940s to the 1960s, has become insubstantial as a result of the population explosion. For example, the municipality of San Juan Chamula had 16 000 inhabitants in 1940 but now has 87 000 inhabitants distributed in more than one hundred localities which are organized in agencies that have conflicting relations with the municipal authority.

From an economic perspective, the importance of the forms of social property in the countryside is worth highlighting. In his reference work, Arturo Warman considers that, in the year 2000, around 854 000 indigenous “comuneros” (or coproprietors) and “ejidatarios” resided and worked in 5 632 ejidos and agrarian communities (^{Warman, 2003}).³ These productive units do not operate in a vacuum. As Aguirre Beltrán demonstrated, the economy of the communities is closely tied to the commercial centers wherein they sell their production and stock up on foreign goods. Therefore, these are part of broader socio-economic systems, in the image of the famous “refuge regions” (^{Aguirre Beltrán, 1967}).

Accordingly, to consider different complementary levels of analysis is useful to identify communitarian policy instruments among the region, the municipality and the locality. Even though they do not perfectly correspond to the “community” level, the sections delimited by the Federal Electoral Institute (IFE) offer advantages for the study of indigenous electoral behavior. These were designed to distribute the population in a balanced manner, so they are relatively homogeneous in their demographic size. The 3 339 predominantly indigenous sections located in the rural area have between 532 (in 1991) and 1200 registered voters (in 2018) on average. These sections, then, make it possible to capture the internal political diversity of the municipalities, the differences and the tensions between the peripheral agencies and the municipal authorities that usually manage public resources. INEGI added the data from the 2010 Census at this constitutive level of Mexican political geography, so we have many socio-demographic data that can be contrasted with electoral behavior on this scale (^{INEGI-IFE, 2012}).

To begin with, I created a dichotomous variable that captures the 3339 predominantly indigenous sections and provides a first approximation to the specificity of the electoral behaviors these microregions comprise, in contrast to the predominantly mestizo sections. As can be seen in Table 2, 65 per cent of all Mexican indigenous populations reside in them, representing on average 85 per cent of indigenous-district population.

TABLE 2 Sectional distribution of the population by percentage speakers of indigenous languages (2010) 

ILS (intervals) (%)	Number of sections	Population older than three years old (2010)	ILS older than three years old (2010)	ILS (%)	ILS Total (%)	Total ILS (accumulated) (%)
90-100	1 692	2 759 497	2 674 958	96.9	39.0	39.0
65-90	1 046	1 593 012	1 260 701	79.1	18.4	57.3
50-65	601	906 503	521 357	57.5	7.6	64.9
50-100	3 339	5 259 012	4 457 016	84.8	64.9	64 .9
40-50	475	748 147	337 028	45.0	4.9	69.8
30-40	574	935 288	326 278	34.9	4.8	74.6
0-30	62 294	94 536 794	1 744 629	1.8	25.4	100.0
Total	66 682	101 479 241	6 864 951	6.8	100.0

Source: Own elaboration based on the Intercensal Survey 2015 (^{INEGI, 2015}).

When one works at this level, the temptation to commit ecological fallacies must be resisted. Generally, each section contains between two and five rural localities, so sectional averages do not allow inferences to be made for the level of individual voters. To locate these sections in their economic and sociocultural environment, we also distinguished seven large indigenous regions, which comprise 2 409 of the 3 339 predominantly indigenous sections and can be found on Figure 3.

Source: Own elaboration based on ^{INEGI-IFE (2012)}.

FIGURE 3 Geographical distribution of seven predominantly indigenous regions (2010) 

How, then, are ethnic-linguistic affiliations linked to recent dynamics of political representation in the Chamber of Deputies?

A persistent lag in the field of legislative representation

Contrary to a recurring idea, the presence of indigenous deputies in the Congress of the Union is not new. As indicated by an investigation carried out during 2011 and 2012, they occupied at least four seats in the LIV Legislature of the Congress of the Union (1988-1991). However, historical trends do reveal a persistent lag in the area of indigenous legislative representation, despite an ephemeral increase in the LX and LXI Legislatures, elected in 2006 and 2009 (^{Sonnleitner, 2013}).

In 2004, the IFE promoted a reform to create 28 federal single-member districts with more than 40 per cent of the indigenous population. This measure of positive discrimination sought to increase the number of indigenous legislators. In 2006, 18 indigenous deputies were elected (one by proportional representation and 17 by relative majority), maintaining a similar proportion in 2009 with the election of 17 deputies (eleven by relative majority and six by proportional representation). However, the effects of this initiative did not last. In 2012 and 2015, their number was significantly reduced to ten, with a continuous decrease in the number of indigenous deputies elected in single-member districts: this number went from seven to six between both Legislatures, and again to seven in 2018 (Figure 4).

Source: Own elaboration based on ^{Sonnleitner (2013)} and ^{INE (2018)}.

FIGURE 4 Number and percentage of deputies of indigenous origin (1988-2018) 

Counter-intuitively, since 2009 the number of elected legislators of indigenous origin has gone down. To clarify this seemingly paradoxical trend, let us analyze two complementary problems related to the design of the 28 “indigenous” singe-member districts; and the operational definition of who can register as candidates in them.

The dispersion of indigenous populations in 28 single-member districts

Firstly, it is necessary to recognize the existence of a possible technical error in the design of the districts, which could be corrected by means of a more efficient affirmative gerrymandering. During the first reform of 2004/2005, a threshold that was too low (40%) was adopted and the highest estimates of the indigenous population of the National Commission for the Development of Indigenous Peoples (Comisión Nacional para el Desarrollo de los Pueblos Indígenas, CDI) were taken as a benchmark. For this reason, between 2006 and 2015 only fourteen “indigenous” districts had, as a matter of fact, effective majorities and, only in ten of them, two out of every three voters spoke any indigenous language (Table 3). Thus, by seeking to increase the number of districts, indigenous voters were dispersed, weakening the expected effects of affirmative gerrymandering rather than creating comparative advantages for indigenous candidates.

This problem was not solved during the last electoral re-distribution process that took place between 2016-2017. Instead of grouping indigenous communities into predominantly indigenous legislative districts (with thresholds of 50 or 65 per cent of indigenous language speakers, to generate more effective affirmative gerrymandering), the same number of districts (28) was maintained with the same threshold (40%) and the same criteria (population in indigenous households). Therefore, only twelve constituencies were created with more than 50 per cent of indigenous language speakers, of which only four have 65 per cent or more of indigenous language speakers, once again diluting the effective proportion of indigenous populations and the potential comparative advantages for their candidates in sixteen multiethnic districts (Table 3).

The disputed indigenous identity: to be, or not to be an“indigenous candidate”

Secondly, a more complex variable must be considered, related to the ambivalence of the criteria for registering as an indigenous candidate. Who competed, and who was elected in the indigenous districts?

The tensions surrounding the nomination of candidates in the thirteen districts with the highest percentage of indigenous population, revealed the difficulties in defining which of them could be considered “indigenous”. From a legalistic standpoint, the INE adopted flexible criteria, privileged self-affiliation and allowed multiple forms of accrediting community ties, for example with proof of having worked for the benefit of some community. This raised questions about a significant number of candidacies. For instance, the challenges that arose in the Chiapas districts 02 of Bochil and 11 of Las Margaritas, where two high-ranking officials of the government of Manuel Velasco Coello (locally known as “el güero Velasco”) ran for office, are illustrative. Despite the fact that public opinion unanimously perceived them as mestizos, they were elected and confirmed by controversial sentences, based on their “qualified self-affiliation” (^{TEPJF, 2018}).

Beyond the controversial criteria adopted by the electoral authorities, the situational complexity and the inter-subjective, contextual and relational nature of indigenous identities in Mexico became clear. For our research, we managed to collect data on the trajectories of 85 of the 105 candidates that were registered in the 28 districts with more than 40 per cent of indigenous population. In addition to self-affiliation and community collaboration, we considered other complementary criteria (including the use of traditional clothing, speaking an indigenous language, having held office in the community and/or indigenous representative positions) to assess whether these candidacies were presented (and were publicly recognized) as indigenous. In contrast to their gender composition (45% of women and 55% of men), only twenty of these (23.5%) were publicly recognized as indigenous (10 women and 10 men) while 65 were considered as being of mestizo origin. In the thirteen districts where political parties had a formal obligation to present indigenous candidates, 44 candidacies were registered (48% women and 52% men), of which ten had indigenous identity (22.7%) and 26 had mestizo identity (no information was obtained for the remaining eight).

In July 2018, four male, and three female indigenous federal deputies were elected. Five of them competed successfully in one of the twelve predominantly indigenous districts (42%), while the remaining two did so in one of the other sixteen districts with more than 40 per cent indigenous population (13%). Thus, a clear lag persists in terms of indigenous representation, which extends here to gender parity as well, particularly in the thirteen districts with more than 60 per cent of the population in indigenous households, where only four women were elected deputies (31%) against nine elected male legislators (69.2%). Box 1 summarizes the biographical information of these legislators and allows locating the districts where they were elected.

Source: Own elaboration with data from the project “Observando los desafíos de la inclusión democrática en México” (Democracia, Derechos Humanos y Seguridad, El Colegio de México, Electoral Observation Fund 2017-2018, United Nations Development Programme).

BOX 1 Elected federal deputies of indigenous origin (2018) 

These results contrast with the progress made in gender equality. Instead of increasing with the new affirmative action measures, the number of legislators of indigenous origin was reduced, returning to the levels of 2003, before the 28 indigenous districts were created. To understand the persistence of these gaps, the question about the existence of an electoral behavior that is specifically indigenous is in order.

How does the indigenous mexican vote?

To begin with, let’s avoid committing culturalist fallacies. The first of these is to assume that elections are external and of no interest to indigenous peoples. As can be seen in Table 4, this premise has no empirical support. Indeed, until 2006, the 3 339 sections with more than 50 per cent of indigenous language speakers were characterized by lower rates of electoral participation. However, since then the trends have been reversed and, today, they register higher averages than those observed in the mestizo sections.

Another common fallacy assumes that indigenous policy is unanimous and consensual, which is why communities oppose multi-party elections. Without a doubt, in many indigenous communities there is a rejection of traditional parties and a commitment to so-called “uses and customs”. However, this movement focuses on the renewal of municipal elites. In the elections for president and federal deputies, the Effective Number of Electoral Parties (Nepel) does not present substantive differences with the mestizo and the predominantly indigenous sections: after having reached 1.6 in 1991 (single-party context), this index reached the threshold of bipartisanship since 1994 (2.6) and tri-partisanship since 2009. Thus, now it stands at an average of 3.4 (that is, in a format of more than three relevant parties, just like what is observed on average in the mestizo sections).

Let us now look at the composition of the vote in the different multi-ethnic contexts (mixed, majority and almost exclusively indigenous). For this, we analyze the results in the 4 352 electoral sections with more than 30 per cent of indigenous language speakers. To contrast them with the rest of mestizo sections, we distinguish five sub-categories with increasing percentages of indigenous populations: the 568 sections with 30-40 per cent and the 473 sections with 40-50 per cent (where mestizos have strong territorial presence), the 596 sections with 50-65 per cent and the 1 039 sections with between 65-90 percent (where mestizos are visible minorities) and the 1 676 sections with more than 90 per cent of indigenous language speakers (where mestizos are a small minority).

In 2018, Andrés Manuel López Obrador (AMLO) and Ricardo Anaya in the presidential race as well as their respective party platforms, Movimiento Regeneración Nacional (Morena) and the Partido Acción Nacional (PAN) in the legislative race all underperformed in the different types of indigenous sections, while the Partido de la Revolución Democrática (PRD) and the Partido Revolucionario Institucional (PRI) held up better there than in the mestizo sections. This different result is also reflected in the percentages of sections that the different candidates/parties managed to win: AMLO wins first place in 79.6 per cent of the mestizo sections but only achieves this in 61.8 per cent of the eminently indigenous sections; Meade (the PRI candidate), on the other hand, only wins in 5.2 per cent of the mestizo sections but reaches first place in 21 per cent of the indigenous sections. As for Jaime Rodríguez Calderón, also known as “El Bronco”, his votes are clearly concentrated in the mestizo sections. The other parties (particularly PT and PVEM) capture a slightly higher number of votes in multi-ethnic contexts, receiving 23.9 per cent in the sections with more than 90 per cent of indigenous language speakers. These figures confirm what we already mentioned above: despite being characterized by a greater presence of the PRI and the PRD, today the indigenous sections have a partisan diversity that is very similar to that of the mestizo regions (Table 5).

However, it would be premature to conclude that the indigenous electorate is more participatory, and more inclined toward the PRI and/or the PRD than its mestizo counterpart. As we will see below, these differences may well be driven by socio-demographic characteristics.

Neither does the “indigenous” category capture the heterogeneity of situations in which the different indigenous communities of the country live in. For example, let’s analyze the vote in seven regions built upon the predominantly indigenous sections. Rather than a consistent behavior, it is more convenient to speak of a marked diversity of votes that are related to the socio-territorial dynamics of these eminently indigenous regions. Voter turnout fluctuates greatly among them, reaching as little as 56.6 per cent in the Huicot-Tarahumara region, or as much as 84 per cent among the Maya of the Yucatan peninsula. Meade’s success is impressive in both regions (where he wins 53.7% and 42.9% of the sections) and contrasts with his mediocre results in Guerrero, Oaxaca and the rest of mestizo and indigenous sections (Table 6).

In turn, Anaya’s results vary strongly among the indigenous regions of Chiapas and Oaxaca (where the PAN is mostly absent), and those that are located in San Luis, Puebla and Yucatan (where he obtains between 28.2 and 31.3 per cent of the valid vote). The PRD barely receives 2.6 per cent of the vote in the Huicot-Tarahumara region but captures 9.9 per cent in Oaxaca, 9.5 per cent in San Luis and 19.2 per cent in the indigenous Montaña de Guerrero. In the end, Morena is not the exception: López Obrador’s results fluctuate between 29.5 per cent in Yucatan and 64.2 per cent in Oaxaca, differences that are due to the success/failure of his state campaigns and not to ethno-linguistic variables.

It is striking to observe that in some indigenous regions the vote is more fragmented than in the mestizo zone, a result captured by the high number of votes going to other parties. This electoral diversity of the indigenous universe is represented in Figure 5, which synthesizes the political plurality of these seven indigenous regions.

Source: Own elaboration based on ^{INEGI-IFE (2012)} and ^{INE (2018)}.

FIGURE 5 The composition of the vote in seven indigenous regions (legislative elections, 2018) 

Schooling, language and electoral behavior (1991-2018)

Secondly, let’s analyze the electoral results of the elections for federal deputies from 1991 to 2018, distinguishing between the trends of the mestizo and indigenous electoral sections and controlling for the average level of schooling, in order to establish if there were specific electoral patterns in the indigenous territories.

As is well known, in many countries electoral participation increases at a higher level of schooling. In Mexico, this relationship is confirmed although it is weaker than in consolidated democracies (at the section level, the Pearson correlation between the school average and the average of electoral participation from 1991 to 2018 is +0.354). However, there is also a clear negative correlation between schooling and the sectional percentages of speakers of indigenous languages (-0.341). Therefore, to establish whether there is an indigenous vote it is essential to analyze the levels of electoral participation in light of the strong territorial inequalities of schooling.

To do this, we grouped the electoral sections into four categories that synthesize the averages of: the 2 697 predominantly indigenous sections with an average schooling of less than six years (where 87 per cent of the population speaks an indigenous language); the 11 511 mestizo sections with an equivalent level of schooling (in which only 3.8 per cent of the population speaks an indigenous language); the 36 727 sections with six to ten years of schooling (in which 3.6 per cent speak an indigenous language);⁴ as well as the 15 730 mestizo sections with more than ten years of schooling (in which only 1 per cent speaks an indigenous language).

As Figure 6 illustrates, electoral participation in Mexico is highly volatile: it reached 78.3 per cent in the sections with the highest schooling in 1994 and bottomed out with 40.3 per cent in the sections with intermediate schooling in the 2003 elections. It always increases when the legislative elections overlap with the presidential elections (in 1994, 2000, 2006, 2012 and 2018) and is notably weaker in the mid-term elections. Until 2006, participation was higher in the sections with more schooling, especially in the presidential elections. However, from 2006 onward it increased significantly in the sections with low education, regardless of whether they were mestizo or indigenous. Finally, since 2009, electoral participation has been equal, or even higher, in the eminently indigenous sections than in the sections with the highest schooling averages.

FIGURE 6 The electoral participation in the mestizo and indigenous sections (by schooling) 

To complete this analysis, I look at the evolution of party preferences in these same four categories. Since 1991 and throughout the entire period, the mainly indigenous sections (marked with triangles) have been characterized by a much stronger presence of the PRI (Figure 7 a) and a much weaker presence of the PAN (Figure 7 b). The differences are particularly noticeable during the period 1997-2006 and reach impressive levels for the legislative elections of the year 2000, in which the PRI obtained up to 29 percentage points more (and the PAN up to 35 percentage points less) in these predominantly indigenous sections compared to the mestizo sections with more than ten years of schooling.

FIGURE 7 The partisan vote in the mestizo and indigenous sections (by schooling) 

The differences will later remain but considerably reduced. In the case of the PRI, this is due to its across-the-board weakening which is observed in all categories but is more noticeable in the sections with lower levels of schooling. In the case of PAN, on the other hand, the convergence is due to its profound drop in the sections with more education and its relative growth in the sections with less education, as a consequence of the diversification of its electorate. A low degree of vote differentiation is observed as well in the case of the PRD, which, with the exception of 1997 (the election of Cárdenas in the Federal District), is more successful in the indigenous sections (Figure 7 c).

Likewise, the similarity of the trends observed in the mestizo and indigenous sections with low levels of schooling (indicated by asterisks and triangles) is worth noting. This provides a first approximation to the effects of schooling on electoral behavior. These effects are particularly apparent in the case of PRI, which has very similar roots in all the sections with less than six years of schooling, regardless of whether they are mestizo or indigenous. Rather than corresponding to an ethnic or cultural behavior, this difference seems to be related to other factors of economic and socio-demographic nature. Finally, the little vote differentiation of the other parties is striking, whose presence is usually slightly higher in the sections with the highest level of education (with the exception of 2003), but whose growth since 2009 is overwhelming and impressive in the four analytical categories.

This first exploratory exercise indicates that the ethno-linguistic variable does seem to have effects on electoral behavior, although it also calls for further analysis, integrating and controlling for other socio-demographic variables.

Ethnicity or exclusion? The weight of socio-demographic and regional variables

To capture the specific weight of ethnicity without confounding it with the effects of other socio-demographic and territorial factors that may also influence electoral behavior, I ran a series of multi-variable regression models, with the data from the latest 2010 Census added at the level of the 66 682 electoral sections (^{INEGI-IFE, 2012}).

The independent variable of interest is the sectional percentage of speakers of indigenous languages (“pHLI”). Likewise, we used seven binary variables (“Indigenous_”, coded with 1/0) to identify the predominantly indigenous sections of each of the seven regions that we previously distinguished, in order to explore the specificity of indigenous electoral patterns among the Yucatecan Mayans, in the Potosina and Puebla Huasteca, in Chiapas, Oaxaca and Guerrero, as well as in the Huicot-Tarahumara region.

Regarding the socio-demographic control variables, I consider, together with the average years of schooling (“MediaEsc”), the sectional percentages of young people between 15 and 24 years of age (“Joven15a24”), the population residing in another entity in the last five years (“Immigrants”), those who worship the Catholic religion (“Catholic”), those who have access to the ISSSTE (Instituto de Seguridad y Servicios Sociales de los Trabajadores del Estado) or the IMSS (Instituto Mexicano del Seguro Social), as well as the proportion of marginalized homes that have neither electricity, piped water nor drained water (“Sin_ServBas”).

To facilitate the interpretation and the direct comparison of all coefficients with the seven regional dichotomous variables, I standardized each continuous independent variable by subtracting its average and dividing it by two standard deviations (“rs”). This procedure, suggested by Andrew ^{Gelman and Jennifer Hill (2007: 56)}, not only enables centering these variables (which allows the intersection to correspond with the predicted average result when the set of variables are located at their respective mean); by dividing each variable by two standard deviations (instead of one), an increase of an integer unit in that variable then corresponds to a change from one standard deviation below to one standard deviation above its respective average (which is approximately equivalent to the change of a binary variable between the values 0 and 1).⁵

Before standardization, I applied some simple transformations to the variables with biased distributions, so that they could have reasonably normal distributions (this is why we used the square roots “Sq_” of the rates of immigrants, Catholics and beneficiaries of the ISSSTE, as well as the natural logarithm “Ln_” of homes with deficiencies in basic services). Since all the variables are similarly standardized, the differences between the various coefficient scales (which can be directly contrasted with the binary variables) are eliminated.

In this way, the constant represents the predicted average of the independent variable of each model (percentage of electoral participation, vote for PRI, PAN, PRD or other parties), and the coefficients of the standardized regressors (in this case, the control variables) can be interpreted on a scale equivalent to that of the coefficients of the binary variables of interest (in our case, the predominantly indigenous sections in the seven analyzed indigenous regions). As can be seen in Table 7, there are relevant correlations between the different variabls. However, these are not strong enough to pose collinearity problems (the statistics of the reported models never go over 3.8 for FIV values, with a tolerance of less than 0.262 for the schooling average).

Let us now turn to data analysis. Considering that the dependent variables vary in a range between 0-100 and that, in practical terms, their distribution is normal, we use linear regression models. These allow the constants to be interpreted as the predicted averages of the dependent variables when all the continuous independent variables are at their mean level and the binary variable has the value of 0. In turn, the standardized coefficients correspond to the total effect that either an increase of two standard variations of a continuous variable or an increase of a unit of a dichotomous categorical variable, in percentage points of the dependent variable, would have.

As observed in the three models in Table 8, the general average of electoral participation between 1991 and 2018 was 58 per cent in the mestizo sections (as indicated by the constant at the intersection), while PRI obtained 40 per cent of the valid vote during the same period, against 26 per cent for PAN and 16 per cent for PRD. As the first model shows, the predominantly indigenous sections are not characterized by a statistically significant difference in terms of electoral participation, but in these sections PRI does seem to achieve +4.7 percentage points more, in contrast to PAN which captures an average of -5.8 points less than in the mestizo sections (model 1 of Table 8). However, these coefficients change substantially when controlling by other socio-demographic variables, with the indigenous sections acquiring a positive value of +2.1 percentage points for electoral participation and decreasing to less than one percentage point for the distribution of partisan votes. In fact, these other socio-demographic factors are much more relevant to explain the sectional variations in the electoral behavior of Mexicans (model 2 of Table 8).

Electoral participation, for example, is positively associated with higher sectional levels of schooling (+10 points), Catholic population (+3.1 points) and homes without basic services (+2.3 points), while it decreases in the sections with the highest presence of immigrants, young people and beneficiaries of ISSSTE and IMSS. As shown by the corrected r² of models 1 and 2, together these variables explain 20 per cent of the total variance, while the ethno-linguistic variable by itself bears little explanatory power. Likewise, the coefficients of the percentage of indigenous language speakers lose their relevance to explain partisan voting. In the case of PRI, the most relevant variable is clearly the schooling average (-12.1 points), which has an exactly inverse effect on the PAN vote (+12.2 points). The latter party also benefits from the greater presence of Catholics (+4.5 points) and IMSS beneficiaries (+4.4 points), in contrast to the PRD that is more successful in the sections with more ISSSTE beneficiaries (+7.5 points) and with deficiencies in basic services (+3.5 points).

The poor explanatory capacity of the ethno-linguistic variable is largely due to its internal heterogeneity. When the dummy variables of the seven indigenous regions that we previously distinguished are introduced, relevant and statistically significant variations appear. As model 3 of Table 8 illustrates, the Mayan sections of the Yucatan Peninsula are characterized by impressive rates of electoral participation (+11.3 percentage points more than the national average) and by a much stronger presence of PAN (+12.6 points ) and PRI (+7.5 points) compared to PRD (-11.1 percentage points) and the rest of political parties (-9.0 points). Also, the overrepresentation of Partido Acción Nacional in the indigenous sections of the Huasteca Potosina (+8.3) contrasts with the overrepresentation of the PRI in the Huicot-Tarahumara sections (+14.5 points) and of the PRD in Montaña de Guerrero (+19.5 percentage points). This illustrates the great diversity of partisan configurations that coexist within the indigenous universe, where many political worlds fit (Table 8).

I also built other models to check whether the global averages are masking substantive changes over time. Although the coefficients vary slightly in intensity, the core results are robust and consistent with the trends seen in figures 7-10. Let us look, for example, at the coefficients of the same dependent variables for the 1997-2006 period. Unlike model 1, which only captures the bilateral effects of the ethno-linguistic variable (and barely captures between 0.4 and 6.2 per cent of the total variance), the other socio-demographic factors explain between 13.8 per cent and 40.1 per cent of the total variance —making the linguistic variable less relevant. Only turnout is slightly higher in sections with greater percentages of indigenous language speakers.

Indeed, the most important variable is clearly the sectional average of years of schooling, with strong positive effects on electoral participation (+12.2 points) and the PAN vote (+14.9 points), as well as negative effects on the PRI vote (-14.6 points). In turn, a higher proportion of Catholics is associated with higher levels of electoral participation (+3.8 points) and PAN voting (+5.2 points), as well as lower rates of votes for PRI (-2.6 points) and PRD (-2.8 points). Likewise, the sectional rates of access to health services help explain the greater success of PRD in the sections with the most beneficiaries of the ISSSTE (+10.6 points), and PAN in the sections with the most beneficiaries of the IMSS (+5.5 percentage points). Finally, the households with more shortages in basic services are positively associated with the PRI vote and the PRD vote and have a negative effect on the PAN vote (-8 percentage points).

But above all, the usefulness of the regional ethno-linguistic variables is confirmed, not so much to increase the explanatory power of the models (in which the corrected r² does not increase substantially) but in order to capture the heterogeneity of their effects (model 3 of Table 9). Once again, the predominantly Mayan sections of the Yucatan Peninsula are characterized by very high rates of electoral participation (+10.6 points), by a very strong presence of PAN (+11.9 points) and PRI (+6.7 points) and by the weakness of PRD (which captures -14.7 points less). Also, the overrepresentation of PAN is confirmed in the indigenous sections of the Huasteca Potosina (+11.7 points), in contrast to the hegemony of PRI in the Huicot-Tarahumara sections (+12.7 points) and PRD in Montaña de Guerrero (+23.4 points). Hence the need to place the different ethno-linguistic communities in their specific territorial contexts.

I close this section with a brief discussion of the most recent federal elections, in order to locate the socio-demographic profile of the winning party, the Movimiento Regeneración Nacional, before and after the electoral tsunami unleashed by AMLO’s third presidential candidacy. While in 2015 Morena was more successful in sections with higher levels of schooling (where it practically doubled its electoral results), its exponential growth canceled the effects of this variable in the 2018 legislative elections, and actually turned the coefficient into the opposite direction for the presidential elections (in which AMLO obtained -8.7 percentage points less in high-schooling sections than in those with low schooling).

Indeed, it is noteworthy that the effects of the same variable remained relatively constant in the case of PAN (which obtained +11.6 points more in the better educated sections), but they weakened considerably in the case of PRI (which only lost -5.1 points in them). Likewise, both Morena and AMLO managed to grow in the sections with higher proportions of young people and of beneficiaries of the ISSSTE and the IMSS, whose preferences used to benefit PRD and PAN, respectively. The only type of sections in which Morena’s penetration was more limited corresponds to the proportion of Catholics (-9.8 points), which was more favorable to the PAN vote (+4.5 percentage points) and, to a lesser extent, to PRI (+2.7 points). Therefore, the 2018 electoral tsunami really blurred the main socio-demographic cleavages that had been structuring Mexican electoral policy since the 1990s.

Regarding the patterns in the different indigenous regions, AMLO only obtained a substantive advantage in the predominantly indigenous sections of Oaxaca (+15.1 points) and Guerrero (+8 points), while registering much lower results in the Huasteca Potosina (-9.1 points) and Poblana (-4.6 points), in the Selva Lacandona and Los Altos de Chiapas (-9 points) but, above all, in the Huicot-Tarahumara (-20.7 points) and Mayan sections of Yucatan (-21.5 points). In effect, the latter remained loyal to PRI and PAN, in the same way that the indigenous sections of Guerrero continued to vote in a greater proportion for PRD (+11.7 percentage points). In any case, it is striking that the vote as a whole in these regions did not favor López Obrador to a greater extent, who obtained a negative electoral balance in the indigenous sections. Finally, indigenous turnout further confirmed the great internal heterogeneity of the linguistic factor, with impressive differences between the Yucatecan Mayans (+14.7 points) and the Tarahumaras-Huicot (-11.1 points).

In short, the explanatory capacity of the socio-demographic factors we use here as control variables is much more relevant and robust to capture the electoral patterns of Mexicans than the percentage of indigenous language speakers. Contrary to a recurring but erroneous myth, turnout rates are significantly higher on average in indigenous sections. However, there is no common pattern of electoral behavior among them. Rather, highly participatory indigenous regions (the Mayans of Yucatan) and highly abstentionists (the Tarahumaras-Huicot or the indigenous people of Guerrero) co-exist, which challenges the idea of a homogeneous or unified indigenous electoral body.

This is even truer when the partisan orientation of the vote is analyzed. As the multivariate regression models show, speaking an indigenous language does not have much explanatory power. On the other hand, strongly differentiated behaviors are observed among the different ethnic-linguistic regions (with markedly PAN orientations among the Yucatecan Mayas; PRD among the Mixtecos, Nahuas, Tlapanecos and Guerrero Amuzgos; or PRI orientations among the northern Tarahumara-Huicot). This illustrates the great diversity of electoral patterns within the different indigenous territories. This heterogeneity refers to geographical and historical specificities that must be recognized, located and explored. Therefore, it is of little use to speak of “an indigenous vote”, and far more appropriate to think of different regions with differentiated electoral patterns.

A plural universe of indigenous worlds: four challenges to improve political inclusion

The pluralism of the indigenous world is worth highlighting. In 2018, electoral participation was higher in almost all of its regions, where party systems are just as fragmented as in the rest of the country. Rather than one vote, it is more convenient to speak of various indigenous votes. As a whole, the predominantly indigenous sections are not characterized by any specific political behavior. And, when the electoral effects of the percentage of indigenous language speakers in the section are compared with the electoral effects of other socio-demographic factors, the latter are usually much more relevant (particularly, the level of education in the section). Certainly, the average level of electoral participation does end up being slightly higher in indigenous sections than in mestizo sections. However, this is not the case for the sectional distribution of the PRI vote, the PAN vote, or the PRD vote, the variation of which is mostly explained by educational levels, access to health services, or lack of basic services, as well as by the sectional proportion of young people, immigrants or Catholics.

This is related to the internal heterogeneity of the ethno-linguistic category. This contains an array of populations with strongly differentiated and sometimes diametrically opposed behaviors, which must be studied in their specific socio-territorial contexts. For the 1991-2018 period, the electoral participation averages varied up to 18 percentage points between the highly participatory Mayan sections of the Yucatan peninsula and the abstentionist sections of Montaña of Guerrero, while the differences between the over-/under-representation of the main parties reach up to 24.6 percentage points for PAN, 25.3 points for PRI and 30 points in the case of PRD, with coefficients twice as high as the level of education in the section (the socio-demographic variable of greatest weight).

These findings have important implications for the political inclusion of indigenous populations. Contrary to the dualistic image that homogenizes them and conceals their internal diversity, the careful study of electoral results reveals a wide rainbow of political preferences that coexist within indigenous territories. This means that indigenous identities must be placed in different geographic, demographic, economic and sociocultural contexts. We have before us a highly diverse group of populations that may inhabit as indisputable majorities in exclusively indigenous communities, or migrate and live in multicultural, mixed or mestizo environments where they become more or less (in)visible minorities.

The design of public policies that promote greater political inclusion of indigenous populations must start from the recognition of their internal diversity. A first challenge concerns the relevant level in which political representation is organized: is it necessary to have specific legal instruments for the 28,000 localities, for the 565 municipalities, for the 23 regions or for the ten entities where most of the indigenous communities live?

As aforementioned, the demographic heterogeneity and the territorial distribution of the ethno-linguistic groups in Mexico invite us to rethink indigenous political representation in multicultural territorial contexts. Despite its geographic concentration, only 57 per cent of indigenous language speakers reside in sections with more than 65 per cent of indigenous language speakers, while 35 per cent of them live in sections with less than 50 per cent, and 25 per cent in sections with less than 30 per cent of indigenous language speakers. Thus, a substantial part of these populations lives in culturally mixed contexts or in eminently mestizo contexts. At what level and through what tools, then, should we design public policy? And what type of political representation for what type of indigenous sectors should be considered?

For the time being, Mexican reforms have resorted to, above all, majority mechanisms of affirmative gerrymandering. The redistricting of 2004-2005 and 2016-2017 designed 28 districts with a deliberate ethno-linguistic bias, while the affirmative action measures launched in 2017 sought to force the nomination of indigenous candidates in at least thirteen of those districts. A first step would be to recognize that the issue cannot be solved solely by these means and that complementary strategies must be used, with a logic of proportional quotas or reserved candidacies.

For example, a sixth plurinominal federal constituency could be created, reserved for indigenous candidates, as was done in Colombia to promote the legislative representation of indigenous and Afro-mestizo populations. Another alternative is to create incentives for the parties themselves to include indigenous candidates in their proportional representation lists, eventually including quota systems in entities or regions with strong indigenous presence. Likewise, the successful experience of “affirmative malapportionment” in Panama is worth mentioning, where the creation of constituencies with two or three seats in the indigenous regions allows them to be over-represented in relation to the mestizo single-member districts (^{Sonnleitner, 2010}). In any case, the fact that Mexico has a mixed representation system could be positively exploited. Electoral engineering opens up many possibilities to combine majority mechanisms (for communities residing in indigenous territories) and proportional mechanisms (for those residing in mixed or mestizo contexts).

Thirdly, the challenge of socio-economic marginalization is worth stressing. Social exclusion does not only afflict indigenous populations but affect broad sectors of the mestizo population as well. However, most indigenous communities live in contexts of high marginalization or extreme poverty, which hinders their political inclusion. Their representation cannot ignore the conditions of material inequality in which those who aspire to public office compete.

Finally, one last challenge must not be dismissed, that is, the counterproductive effects that positive discrimination can cause: formally recognizing indigenous identities as subjects of exclusive public policy entails the risk of creating unexpected dynamics of exclusion and resentment among non-indigenous sectors. This could feed or reproduce old and new practices of paternalism and racism under the guise of benevolent discourses that could break the existing consensus on the legitimacy of indigenous political inclusion.