«By Celeste Nicole Henrickson A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in ...»
Kroeber (1931) constructed a distribution of cultural traits for the Guaycura, Cochimí, and to a lesser extent, the Pericú (as well as several other non-peninsular groups) based on the Jesuit missionary accounts of Johann Jacob Baegert, as well as research by Miguel Venegas and Francisco Xavier Clavijero (1789) to determine their relationship with the Seri/Comcáac. Kroeber tentatively suggested: 1) the Guaicura had greater cultural affinity to the Seri/Comcáac than did the Cochimí; and 2) the cultural groups of lower California as a whole were less culturally similar to Seri/Comcáac than the Gila Pima or Walapai-Havasupai. His analysis consisted of adding the number of traits each group shared with the Seri/Comcáac subtracted from the number of traits they did not share; however, he did not use the same cultural-trait distribution to measure all relationships. Instead, the number and kind of traits used to determine relationships differed for each pair-wise grouping with the Seri/Comcáac.
As a result, there is no way to meaningfully interpret the nature and magnitude of the differences between groups as each was measured on a separate metric. When inferences of group similarity are based on the presence or absence of cultural traits and cultural traits are missing for some groups, both Type I (incorrectly identifying a difference exists between groups when none is present) and Type II (incorrectly identifying a similarity exists between groups when in fact they are different) errors are inflated, causing a distorted picture of relationships to emerge. Although a valiant first attempt, the paucity of data and the lack of statistical controls render any interpretation difficult. Bowen (1976:102) echoes similar concerns about the interpretation of Kroeber’s data, but suggests the overall pattern is likely correct. Additionally, he suggested Kroeber’s trait list be revised and expanded to discern relationships between BCS groups and the Seri/Comcáac. To our knowledge, no one has proceeded with this task.
Content analysis is a research methodology that can render the anecdotal nature of historical documents into systematic cultural trait distribution lists. Once constructed, cultural trait data allows hypotheses to be tested about macro-level social interactions between groups (Ryan and Bernard 2000). Reliability and hierarchical 14 cluster analysis are then utilized to determine cultural relationships within BCS populations and between them to the Seri/Comcáac. Due to the dearth of theory and data for predicting relationships between cultural groups in southern BCS, our analysis largely is exploratory. Based on our analyses it appears: 1) the historic Guaycura and southern Cochimí were highly culturally similar; 2) the degree to which the Guaycura and southern Cochimí shared cultural traits was much greater than any other grouping;
3) the Seri/Comcáac form a distant but natural grouping with the Guaycura and southern Cochimí; and 4) the Pericú are culturally distinct.
Methods This study applies classical content analysis (Ryan and Bernard 2000) to investigate relationships between three historical BCS cultural groups: the Pericú, Guaycura, and southern Cochimí; as well as the Seri/Comcáac of the Gulf of California and Sonora Mexico (Figure 1). Content analysis uses messages (e.g., texts), rather than behavior or artifacts, as the unit of study (Neuendorf 2002). Although content analysis consists of a range of techniques, the basic premise is the same; researchers convert qualitative texts into quantitative data, which can be used to test relational hypotheses (Ryan and Bernard 2000). Content analysis requires: 1) selecting texts for analysis; 2) defining the variables to be coded; 3) applying those codes systematically to a set of texts; 4) testing the reliability of coders when more than one is present; 5) creation of a unit-of-analysisby-trait matrix from the texts and codes; and finally, 6) hypothesis testing using statistical methods (Bernard 2002). The traits examined include institutional structures and ethnic markers identified by historic explorers and Jesuit missionaries, as well as cultural anthropologists, archaeologists, and historians who have examined or translated these ethnohistoric documents.
There are multiple ways to derive lists of cultural traits. Archaeologists derive them from the material record; cultural anthropologists generally use ethnographies.
Cross-cultural anthropologists commonly employ distribution lists of cultural traits to test relational hypothesis about the ecological, historical, or social correlates of human behavior (e.g., Barry and Schlegel 1980). There are drawbacks to this type research (also termed holocultural research). For example, culture trait lists generally are based on a few descriptive sentences about the presence or absence of a particular trait in a particular culture at one given time. Additionally, coding the presence or absence of a trait for a cultural group based on limited informants masks all of the variability that often exists within a culture (Hewlett and Macfarlan 2010), particularly age, gender, and status. Given these limitations, holocultural research can be useful for determining broad patterns between cultures.
Two reviewers derived a list of cultural traits based on four classes of source material: 1) translated historical documents related to explorations of Baja California spanning the period of A.D. 1539 – 1721 (Alarcón 1992; Atondo y Antillón and Kino 1992; Ascensión 1992; Cardona 1992; Cooke 1992; Lucenilla 1992; Nava 1992; Ortega 1992; Porter y Casanate 1992; Shevlocke 1992; Ulloa 1992a, 1992b; Vizcaíno 1992a, 1992b); 2) translated Jesuit missionary accounts of Baja California Sur groups spanning the period of A.D. 1683-1768 (Baegert 1952; Barco 1981; Burrus 1984; Nunis 1982); 3) Seri/Comáac ethnographic accounts (Bowen 2000; Felger and Moser 1991; Kroeber 1931; McGee 1898) and archaeological research (Bowen 1976); and 4) peer reviewed academic research pertaining to historic BCS (Aschmann 1967; Heizer and Massey 1953;
Kroeber 1931; Laylander 2007; Massey 1947, 1949, 1961, 1966; Mathes 1992, 2006). As an initial exploration of a method with a time consuming research process, I chose not to 15 include information from groups living north of the 28th parallel in the peninsula. The logic is this approximates the northern boundary of southern Cochimí language (Laylander 1997; Mixco 2006) and groups below this boundary had definite, documented contact with Europeans since A.D. 1539 (Mathes 1981). The Cora and Monqui are excluded because insufficient data existed for a separate trait analysis and their group and linguistic affiliation is ambiguous.
Assigning Traits to Groups A list of 88 candidate cultural traits was derived from initial readings of Baegert (1952), Barco (1981), Burrus (1984), and Nunis (1982). These traits largely describe Guaycura and southern Cochimí culture, and to a lesser extent, the Pericú. Cultural traits pertaining to the Seri/Comcáac were easily derived as trained ethnographers and linguists have researched these groups since the pioneering work of McGee (1898).
Pericú cultural traits were the most difficult to code due to a lack of data; however, historic explorer accounts spanning the period of A.D. 1537-1712 were vital for reconstruction. This process caused our final data set to be truncated to 51 cultural traits, grouped into five categories: 1) Male Headdress; 2) Female dress; 3) Religious practices/marriage; 4) Child Carrying Devices; and 5) Technology (Appendix II).
A presence/absence dichotomizing method was used to assign cultural traits to groups. To avoid sample bias, a trait was used only if sufficient information existed for all four groups. The rationale is, if an account identified a trait for one group with no information recorded concerning the other groups, marking an absence of this trait could inflate the similarity of the other groups when statistical analyses are run. Some traits were recorded as “not present” when an alternative version was present and the author made no claim about the trait’s absence.
A two-tiered system was utilized to reconcile instances where accounts/reviewers differed on presence/absence of a trait. If a trait was suggested by one account to be absent, but another recorded its occurrence, the author who reported the presence was chosen. The rationale is it is easier to mistakenly attribute a trait’s absence than its presence. Secondly, authority was deferred to accounts where the author spent a greater deal of time with a group than to authors who never visited the peninsula or visited briefly. The rationale is these authors should have greater cultural knowledge of the groups they discuss.
Results Due to human error (e.g., incorrect reading of text or data coding), it is important to evaluate inter-coder agreement or reliability (Ryan and Bernard 2000). Reliability concerns whether a measuring procedure yields the same results on multiple trials (Carmine and Zeller 1982) and is evidence that a coded theme has some external validity (i.e., it is not a figment of the researchers imagination) (Ryan and Bernard 2000).
As such, reliability analysis was used to determine accuracy between raters.
Conventions in reliability analysis are varied; however, many authors agree coefficients greater than 0.7 are sufficient for exploratory research to perform subsequent analyses (Landis and Kosh 1977). High inter-rater agreement was achieved for traits assigned to the four cultural groups (Pericú: Cohen’s K=0.95, N=51, p0.001; Guaycura: Cohen’s K=0.8, N=51, p0.001; southern Cochimí: Cohen’s K=0.92, N=51, p0.001;
Seri/Comcáac: Cohen’s K=0.92, N=51, p0.001). When disagreements occurred on a trait’s proper coding, primary source material was reviewed and the appropriate scheme determined through consensus. Thus, the consensus building process 16 eventually resulted in perfect agreement between raters for all traits for all four cultures.
Reliability and hierarchical cluster analyses were employed to determine data structure. Reliability analysis determines a set of items’ internal consistency when measured with Cronbach’s alpha (Vogt 2005). When items are cultural traits, reliability analysis determines the extent to which groups share a culture. High reliability coefficients (e.g., 0.7) indicate that groups share a common culture. Low reliability coefficients indicate groups are culturally distinct from one another. A low reliability coefficient was derived when all four cultural groups were examined simultaneously (Cronbach’s α=0.40; N=51). A second set of reliability analyses were run examining three cultures simultaneously, which revealed moderate to extremely low reliability
coefficients (Guaycura-southern Cochimí-Seri: Cronbach’s α=0.47; N=51; PericúGuaycura-southern Cochimí: Cronbach’s α=0.47; N=51; Pericú-Guaycura-Seri:
Cronbach’s α=0.15; N=51; Pericú-southern Cochimí-Seri: Cronbach’s α=0.17; N=51). A final set of reliability analyses examined pairs only. High internal reliability was reached for the Guaycura and southern Cochimí (Cronbach’s α=0.77; N=51); however, extremely low or negative reliability coefficients were derived for all other pair-wise groupings (Table 2). Negative reliability coefficients are indicative of small sample sizes or the evaluation of multiple constructs (Krus and Helmstadter 1993) – i.e. different cultures. Although the sample is moderately small, it appears multiple cultures were examined simultaneously. Given the high cultural trait agreement between southern Cochimí and Guaycura, the additional constructs being evaluated are the Pericú of the Cape Region and Seri/Comcáac cultures of mainland Mexico.
Due to the moderate reliability estimates for the Guaycura-southern CochimíSeri and Guaycura-southern Cochimí-Pericú groupings, further exploration is needed to determine whether deeper structures exist within the data. Researchers, including anthropologists (e.g., Maxwell et al. 2002), employ cluster analysis when a set of objects’ natural classification is unknown and taxonomic order is desired (Aldenderfer and Blashfield 1984). Hierarchical cluster analysis is one clustering technique that places single entities into increasingly homogeneous groupings using an iterative process.
Although standards vary, many agree hierarchical cluster analysis is a preferred clustering method for small sample sizes (e.g., 250 cases), with a minimum requirement of no less than 2k cases (k=number of variables) (Dolnicar 2002).
Hierarchical clustering requires a similarity metric to assess distances between groups and a link-function to hierarchically organize them. It is vital to have a justification for selecting one similarity metric and one link-function over others, as output is determined by these choices (Aldenderfer and Blashfield 1984). In this case a Phi 4-point correlation similarity metric and a within-groups link function were utilized. The Phi 4point correlation procedure was selected over other binary data similarity measurements because of its ease of interpretability (it is equal to the Pearson product moment correlation coefficient for binary data) and it gives equal weight to the joint presence 17 and absence of traits to calculate similarity. Because traits were selected where the joint absence of a trait was equally meaningful as their presence, this metric is more appropriate than those that exclude joint absences from computation (Aldenderfer and Blashfield 1984). The within-groups link function was selected because it was designed for the specific purpose of determining homogeneity within clusters by examining both inter- and intra-cluster pairs (Garson 2009). This resulted in two classifications: 1) the geographically adjacent southern Cochimí and Guaycura of the south-central peninsula form a distant yet single group with the Seri/Comcáac of mainland Mexico; while 2) the Pericú of the southern peninsular tip were isolated. Identical results were obtained using other similarity metrics (i.e., Lambda, Anderberg’s D, and Yule’s Q).