*Statistical Analyzes*

Preliminary analysis included a measurement error study. All data collection and measurements were conducted by a single operator to prevent the effects on interobserver error. Measurement error was investigated by using an analysis of variance, where measurement error was calculated as the proportion of the mean-squared differences between replicates relative to the total between-group variation [14]. The subsample (*n* = 15) of known hemisphere siding were measured on two separate occasions and measurement error (ME) calculated as % ME = 100 × MS (within)/MS (within) + MS (among). Measurement error ranged from 0% to 3% (results not shown), and with this low measurement error, we considered intraobserver error had a very minimal effect on further analyzes.

Canonical Correspondence Analysis (CCA) initially examined the potential association between the four metrics: occipital height, both left and right (in mm) and width, both left and right (in mm), and the presence or absence of a left, right, or no occipital bridge (Table 1). CCA is particularly suited to datasets where quantitative variables and presence/absence variables are common, such as ecological datasets [15]. Only recently has this been applied to brain evolution, specifically quantitative variables, and the presence/absence of sulcal patterns [16]. CCA allows a comparison analysis, directly testing a priori hypotheses emphasizing the variance of *Y* that is related to *X*, and where CCA combines the properties of both ordination and regression analyses to produce ordinations of *Y* that are linearly constrained to *X* [15]. Correlation analysis then tested the strength of the potential correlation between two or more variables using the most common correlation statistic (Pearson's *r* correlation coefficient), with a two-tailed significance that the variables were uncorrelated and a Monte Carlo permutation (using 9999 iterations) [17].


**Table 1.** Occipital lobe measurements and bridging pattern type.

**Table 1.** *Cont.*


1 All numbered measurements in left (L) and right (R) height and width in mm. 2 Presence (Y), absence (N), or Both (B) of a visible bridging gyrus. \* Indicates the subsample of individuals with known siding.

To estimate the uncertainty due to unknown hemisphere siding, a subsample (*n* = 15) where the hemisphere siding was known (left and right) was examined with Bivariate ordinary least-squares (OLS) regression to test the strength of association between each of the four variables and occipital lobe side (left and right hemisphere). For regression purposes, and to linearize scaling relationships [18], each variable was converted (from mm) into natural logarithmic units (base *e*) and a 95% confidence interval fitted to the log–log regressions.

Predicted height and width from both hemispheres was calculated using prediction equations provided by the bivariate OLS regression models, where y = (*a* × log[x] + *b*). The reliability of the predictions was calculated as the percentage of prediction errors (PPE), where PPE = (predicted − observed)/predicted × 100). PPE calculates the uncertainty in an estimate relative to its size [19]. Prediction reliability was determined by applying a bracket of uncertainty produced by the standard error (s.e.) from the bivariate OLS regression models calculating the upper and lower estimates for predicted height or width for each specimen relative to its size, wherey=(*a* × log[x] + *b* ± s.e). This maintained any inherent differences between each variable allowing for changes in the range of uncertainty, where each variable is associated with differences in the standard error [20]. All statistical analyses were conducted in *Past 4.0* [21].
