PHASE BY PHASE TRENDS AT EL UJUXTE

Once El Ujuxte blades had been assigned to these phases, we could see that blade-sample size varied widely among the three phases. Of 1,260 blades in the study, only 10 are from the earliest, Caramelo phase; 37 are from the Cataluña phase. This leaves the preponderance of blades—1,213, or 96.3%—in the latest, Pitahaya phase.

Notwithstanding this skewed distribution of blades across the three phases, there are consistent trends in the data of Figures 2, 3, 4, and 5. In all four bar charts, mean dimensions and weights decrease regularly through time from phase to phase, with changes in lengths and weights being more pronounced than those in widths and thicknesses.

Mean length by phase.

Mean width by phase.

Mean thickness by phase.

Mean weight by phase.

Are differences among the phases for these four size variables collectively significant? While t-tests could be used to test significance for each pair of means, the problem with separate t-tests is that results are presented as independent outcomes, while clearly they are not. So if we find weight to be different and length to be different, we are really presenting the same result twice if length and weight are highly correlated. To explain the chi-square test employed, we have to remember two things. First, the chi-square distribution with p degrees of freedom is the distribution of the sum of squares of p standard normal variables. Second, the central limit theorem says that under very general conditions, averages (means) are approximately normally distributed. We apply this in the following way. Figures 2, 3, 4, and 5 show one set of means for each phase. These means will be approximately normally distributed, which means that the difference between the rows will be approximately normal, as well. To compute the chi-square, we cannot simply take the sum of squares of the differences between rows, because columns (variables) are not independent. Length, weight, and so on are (pretty strongly) correlated. We use the actual correlations to weight the sum of squares of four correlated normal random variables in such a way that it becomes approximately the sum of squares of four independent normals—that is, it becomes approximately chi-square distributed. Testing the hypothesis that the rows of means are equal, and taking the correlation of variables into account, between the Pitahaya phase and the Cataluña phase, the chi-square is 7.61, with P < .04; between the Cataluña phase and the Caramelo phase, the chi-square is 5.74, with P < .08; and between the Pitahaya phase and the Caramelo phase, the chi-square is 16.03, significant at p < .001.

Another way of approaching the same problem is with a discriminant analysis (Figure 6). Here we make a weighted sum of the four variables in such a way that (a) the weighted sum has variance one; and (b) the means of the three groups (phases) are as far apart as possible. Thus, we maximize the between-group variance relative to the total variance. It turns out that the weighted sum (linear combination) we make accounts for 95% of the between-group variance—that is, there is effectively only one dimension (size) in which the three groups differ. Figure 6 shows the distribution of blades in terms of this new “size” variable for the three phases, with “smallness” represented by positive values to the right and “largeness” by negative values to the left. One can see an increasing shift from large blades in the earliest group (group 3, or Caramelo phase) to small blades in group 1 of the Pitahaya phase.

Discriminant analysis.

Also, mean length-to-width ratios decrease regularly (Figure 7). In sum, blades gradually become both smaller and squarer through time. Given the earlier research at La Blanca, these regular trends of both decreasing blade size and changing proportions had been hypothesized. And so we find a parallel trend and statistical support for the idea that there is some tendency on the southern coast during Preclassic times for blades gradually to become smaller and squarer through the sequence.

Mean length:width ratio by phase.

TRENDS WITHIN THE PITAHAYA PHASE

If this were a pervasive trend, then it seemed predictable that similar shifts would be found not just from phase to phase but also continuing through El Ujuxte deposits of the Pitahaya phase, which produced 96% of all blades in the study.

We focused, then, on a large Pitahaya subset of 1,196 blades. These came from 12 excavation units with deep deposits, averaging about 3.0 m per 2 × 2 m square, and we divided the levels arbitrarily into those above and below 2 m. This produced two large, roughly equal samples of blades, 611 above 2 m (levels 1–20), and 585 below. Much larger Pitahaya samples allowed direct use of the chi-square test, so for dimensions, weights, and length-to-width ratios, we used means to divide the blades into large and small fractions. Along with cross-cutting sampling from upper and lower deposits, this resulted in two-way, four-cell tables. Blade thicknesses and widths (Table 1) vary little by macro-level when viewed in this way. However, lengths decrease significantly from lower to upper Pitahaya deposits (Table 2), as do weights and length-to-width ratios. These trends parallel findings described earlier among the three El Ujuxte phases.

Blade width by depth, Pitahaya phase

Blade length by depth, Pitahaya phase

CORRESPONDENCE ANALYSIS: DISTRIBUTION OF SAMPLES

Although these trends are significant statistically, they are a bit suspect, since blades can sometimes occur concentrated in large quantities. Several large deposits of blades representing different tasks could produce an apparent sequence with little if anything to do with gradual cultural change. There is reason, then, to search for evidence of a pervasive trend within individual excavation units that produced Pitahaya-phase material. This was accomplished through a statistical technique known as correspondence analysis (Greenacre and Blasius 1994). Archaeologists occasionally have employed correspondence analysis in their research (e.g., Bølviken et al. 1982; Clouse 1999); the authors (Nance et al. 2003) recently used this technique in a study of ceramics from Iximché.

Working with the same large Pitahaya subset, we made a summary dataset with information about 31 samples of blades, each sample coming from a group of contiguous 10 cm levels of a single excavation unit. The idea was to maximize sample size and still preserve evidence of stratigraphic change. The average sample contained data on 36 blades and represented between 11 and 12 levels of an individual unit. This summary dataset contained information not only on dimensions and weights, but also on edge wear, blade modification, and other variables, including blade proportions relative to available debitage, which we had also summarized. It did not, however, contain information on provenience. That is, level and unit designations were not incorporated.

In a correspondence analysis, distances among samples are computed using chi-square statistics. One can imagine three samples and three chi-square tests contrasting each sample to the other two. For a study of, say, five variables, each test would involve a ten-cell contingency table. The results for three samples could be represented on a two-dimensional grid. Sample loci close together on the grid would resemble one another in variable frequency proportions, since the distance between them would represent a small chi-square value. With many samples, the procedure is the same, except that it is impossible to conceptualize visually the multiplicity of relationships. The correspondence-analysis statistic, however, reduces all of the chi-square distances among all the samples onto a two-dimensional grid in what can be considered a best possible fit.

The same analysis also computes chi-square distances among the variables. In the earlier example, for three samples and five variables, we would generate 10 chi-square tests, each for a 6 cell contingency table, contrasting the frequencies for two variables as distributed within three samples. Both sample and variable loci can be depicted on the same plot (or joint plot). Variable loci situated near each other have similar distributions among the samples, and variable loci are situated near samples where they are highly represented.

In the study at hand, we generated a single correspondence analysis for 31 samples and 16 variables (represented by 46 constituent attributes), so the resulting joint plot contains 77 labeled points or loci. To see the results with some clarity, however, only selected loci are depicted in three figures (Figures 8, 9, 11), all taken from the same analysis.

Figure 8 shows all 31 samples projected on a two-dimensional plot. Also in Figure 8, samples from nine of the individual units are shown connected with a line, the arrow pointing from the upper to the lower sample. Based on stratigraphic evidence, then, there is a time line running through the chart from upper right to lower left. This trend is evident for nine of the 12 squares represented in the analysis. In short, there is evidence of a blade sequence running through the 200 years of Pitahaya occupation.

Correspondence analysis, sample loci.

CORRESPONDENCE ANALYSIS: DISTRIBUTION OF VARIABLES

Figure 9 depicts sample loci for only excavation Unit 17-2, the three samples isolated in the upper left-hand corner of Figure 8. The three samples for this unit do not line up chronologically—that is, diagonally across the plot—as do those for the nine units shown in Figure 8. Also included in Figure 9 are loci for four of the 16 variables that figured in the analysis. These have to do with size: length, width, thickness, and weight. The constituent attributes of each variable are plotted, and each of these summarizes frequencies of blades with above- and below-average measurements for the size variable in question. In other words, on the chart, each size variable is represented by two attributes. It so happens that for all four variables, the attribute locus having the higher Y value on the chart represents the smaller size attribute, and the lower Y value represents the larger. For example, the width variable is summarized through two attributes: below-average width and above-average width. The attribute above-average width is represented by an X below the Y = −.5 grid line in Figure 9. The locus for below-average width is represented by an X above the Y = .5 grid line.

Correspondence analysis, variable loci.

These data in Figure 9 allow us to examine the other perspective of Pitahaya blade distributions—horizontal rather than vertical and synchronic rather than diachronic. Can we discern subassemblages that might reflect differing patterns of blade use across the site? Given that the positions of the three sample loci of Unit 17-2 are in close proximity to all four below-average size attributes (Figure 9), we would anticipate that the 180 blades from these samples, combined, would be quite small compared with blades from the 28 other Pitihaya samples. This turns out to be the case. Not only are lengths (Table 3) and weights lower for these “outlier” samples; but widths (Table 4) and thicknesses are lower, as well, and these differences are highly significant. We do not seem to be dealing here simply with a late portion of the sequence—for example, length-to-width ratios vary little when comparing these two groups of samples. The excavation unit in question was situated in a level area near the bases of two mounds (Mounds 36 and 38), possibly where trash was dumped or accumulated (Love, personal communication 2004). For some reason, those working at this locality required and obtained delicate blades produced from small cores.

Outlier versus other samples by mean length: Correspondence analysis samples, Pitahaya phase.

Outlier versus other samples by mean width: Correspondence analysis samples, Pitahaya phase.

It can be seen in Figure 9 that the weight-dimension attributes cluster in four pairs: weight + (i.e., above average) and length +; weight − and length −; thick + and width +; and thick − and width −. In other words, blade length co-varies closely with weight among the samples, and blade width co-varies with thickness. In Figure 8, the nine line segments, each representing a different excavation unit, all have positive slopes, and such would be the case for a line segment approximately connecting the pairs weight +, length + to weight −, length −. This positive diagonal represents time to some extent, and we know that blade lengths (Table 2) and weights become smaller through the Pitahaya phase. In contradistinction, a line segment connecting thick +, width + to thick −, width − would have a negative slope, and there is no clear evidence that blade width (Table 1) or thickness changes appreciably through Pitahaya levels. Of course, one can see two general clusters, as well—small blade dimensions and weights, on the one hand, and large dimensions and weights, on the other—separated along the Y axis. As discussed earlier for interphase differences, all size variables tend to co-vary. However, based on available Pitahaya evidence, it seems that lengths and weights have more to do with time, and widths and thickness have more to do with functional variability across the site. Table 5 summarizes, by Pitahaya sample, lengths (in mm) and weights (in grams) for the nine excavation units with aligned samples in Figure 8. In terms of these dimensions, average blade size generally decreases within these units from lower to upper samples. At the same time, standard deviations in Table 5 are large, indicating size diversity within samples; sample size in most cases is small. Due to the problem of sampling error, this trend does not show up well within individual units on a level-by-level basis. For example, Pearson's correlation coefficients comparing blade lengths and level assignments for each unit (Table 5) are only statistically significant (P < .05) for three of the four excavation units producing at least 50 Pitahaya blades and for none of the other units represented in the table.

Blade weights and lengths for selected Pitahaya samples.

USE WEAR AND BLADE SIZE

Other variable loci in the correspondence analysis figure in this research. Not yet considered is the idea that decreasing blade size through time might have been due to increased modification (retouch) or edge wear. For example, it is certainly plausible that if the incidence of bipolar flaking had increased, this could have caused blade size generally to diminish. However, bipolarly flaked blades, while common enough in the collection (Figure 10e–h), did not increase through the Pitahaya sequence. Proportions of these modified blades actually decrease slightly from 5.5% of blades below level 20 to 4.1% in levels 20 and above. In terms of dimensions and within the same Pitahaya sample, bipolarly flaked blades are significantly below average in width but significantly above average in thickness. It seems likely that thin blades would not have stood up to the percussion-flaking of the bipolar technique.

Blade illustrations.

Figure 11 shows variable loci in the same correspondence analysis having to do with deliberate edge modification: on blade ends (e.g., burins; Figure 10i–j); through bipolar flaking; and on lateral edges (Figure 10k). One other variable included in Figure 11 summarizes edge alteration interpreted to be the result of use—in this case, wear on blade ends (e.g., drill bits; Figure 10l–n). These four variables are each expressed in terms of two attributes, one summarizing frequencies of blades with, versus one for those without, one of these forms of (interpreted) deliberate modification or incidental edge use wear. Figure 11 shows that attribute loci for these variables where evidence for wear or modification was absent (indicated by a minus sign) cluster together in the upper left corner of the chart and are situated near the small blade-size attributes (cf., Figures 9 and 11). The positive attribute loci for these same variables are located below and to the right. The correspondence analysis, then, does not suggest a tendency for blades of small size and those manifesting one or more of these forms of edge alteration to cluster on a sample-by-sample basis, or, likewise, that these attributes tend to be represented together on the same blades.

Correspondence analysis, other variable loci.

As interpreted here, the most pervasive form of edge wear from use on El Ujuxte blades consists of very fine pressure flaking along blade lateral edges (Figure 10a–d). We found that blades showing this form of use wear, instead of being diminished in size, are actually significantly larger in all dimensions compared with those lacking this lateral edge alteration. All together, blades showing marked lateral edge use tend to be large and seem better suited for heavy tasks (cf. Nance and Kirk 1991).

Also, this general form of lateral edge wear does not become more pronounced on blades above level 21 of the Pitahaya sample. Actually, there is a slight though statistically insignificant tendency for pronounced edge wear to occur on Pitahaya blades more frequently in lower levels.

In sum, there is no clear evidence that diminished blade size in upper Pitahaya levels was due either to more intensive use or to an increase in deliberate modification.

EXTRA-SITE COMPARISONS

Detailed comparisons of La Blanca and El Ujuxte data do not reveal a regular sequence from large to small blades when samples from the two sites are interdigitated chronologically. However, there are size decreases at each site. This lack of consistency may be due to small sample sizes and sampling error or possibly, to local histories that may or may not be individual expressions of a regional trend. John Clark (1988:46) reported for the Middle Preclassic site of La Libertad in the central depression of Chiapas a trend “away from a specialized blade industry towards an industry of expediency.” At all three sites, then, during Late or Middle Preclassic times, we see evidence of a deteriorating blade industry.

Another area of comparability has to do with access to blades by people of high rank versus commoners. Clark (1987:280–281), in an assessment of excavated Middle Preclassic sites in highland Chiapas, saw blades as status markers in those communities, with most blades belonging to high-status individuals. Also, investigators working in Vera Cruz have found high concentrations of obsidian or obsidian blades in site centers or ceremonial or elite contexts (Barrett 2003:216; Knight 1999:176–179; Santley 1989:143). Certainly, these observations are compatible with our findings for the Late Preclassic at El Ujuxte. As indicated earlier, the possibly elite residence, Mound 46, produced larger blades than other contexts.

Earlier, we noted that diminished blade size can be viewed as either a positive or a negative development for prehistoric populations. In Europe, blade size appears to decrease regularly through the Upper Paleolithic (Owen 1988:157–158), presumably resulting in the microliths of Mesolithic and Neolithic times. While this could represent a worldwide trend toward smaller, lighter, more efficient tools, we can ask why it shows up so clearly through just 200 years of Maya prehistory.

Laura Kosakowsky et al. (1999) identified extensive trade in Fine Red pottery from the coastal plain of southeastern Guatemala into the adjacent highlands and suggest a contrasting “flow of obsidian” from the highlands in exchange. In their interpretation, this “exchange of finished products … becomes institutionalized in the Classic period,” trade that had “become formalized by well established trading relations now dating back to the Late Preclassic” (Kosakowsky et al. 1999:388). Our own research with obsidian from two sites on the northwest southern coast of Guatemala and that of Clark (1988) at La Libertad in Chiapas suggest more tenuous trade in highland Guatemalan obsidian to surrounding regions during the Middle to Late Preclassic period.

RESUMEN

Se resumió los datos obtenidos sistemáticamente de aproximadamente 1,200 navajas de obsidiana procedentes del sitio El Ujuxte, ubicado cerca de la costa sur de Guatemala, de acuerdo a tres fases cerámicas que Michael Love definió en 2002 para dicho sitio. Es decir, asignamos navajas a las fases con base en el contenido cerámico del contexto en el cual se les descubrió. Se detectó que de fase en fase occurió disminuciones en las dimensiones de las navajas, en los pesos y en las proporciones de largo:ancho. Nosotros habíamos predicho estos resultados al tomar como base los descubrimientos anteriores que realizamos en el sitio La Blanca (Nance y Kirk 1991).

La mayoría de las navajas de El Ujuxte se deviaron de la fase Pitahaya del preclásico tardio. Y aún dentro de sus 200 años de trayectoria, las navajas muestran un cambio notable hacia tamaños más pequeños, medidos en términos de largo y ancho. Esto puede observarse en 9 de las 12 unidades de excavación con buenas muestras de navajas (Cuadro 1). También, encontramos variabilidad a lo largo del sitio durante la ocupación Pitahaya, como aparece en el análisis de las pruebas estadisticas de correspondencia. Una de las unidades de excavación produjo navajas particularmente pequeñas, que no sólo disminuyeron en longitud y peso sino que también tuvieron anchos y grosores menores. En un área del sitio, que fue identificada como una residencia de elite se encontró navajas significativamente más largas que las otras encontradas en el resto del sitio durante le ocupación Pitahaya.

Asimismo, registramos sistemáticamente la evidencia de las huellas de uso y las modificaciones de retoques en estas navajas, y pudimos determinar que los cambios en los tamaños de las navajas no fueron resultado de estos factores. También, al estudiar la morfología de las navajas, determinamos que el cambio en la longitud de las navajas no fue determinada por quebrarlas para producir más herramientas que fueran más cortas. Las densidades de navajas decaen en los depósitos superiores de la fase Pitahaya, y la explicación más probable para todas estas tendencias es que tanto las navajas como la obsidiana fueron cada vez más escasas para los habitantes de El Ujuxte.