Meta-analytic tree models (meta-CART) can be widely used to identify possible (multiple) interaction effects and examine how study characteristics explain the heterogeneity in study effect sizes....Show moreMeta-analytic tree models (meta-CART) can be widely used to identify possible (multiple) interaction effects and examine how study characteristics explain the heterogeneity in study effect sizes. The algorithm partitions individual studies in more homogeneous groups (i.e. terminal nodes) and in each node a summary effect size is estimated with a naive standard error. Unfortunately, tree based method can be unstable. Since the terminal nodes are generated by an algorithm rather than being predetermined, the complex search strategy that generates these naive standard errors fails to distinguish between a constant (global) and a local optimum and can be over-optimistic. In order to acquire more attainable confidence intervals of the summary effect sizes, the current study introduces a new bootstrap calibration approach to meta-CART models. With bootstrapping, the standards errors are re-estimated in order to correct this underestimation of the naive confidence intervals. The present study was conducted to test the performance of the new bootstrap method extensively via a simulation study and aims to provide more knowledge whether the new approach showed better coverage of the established tree models. The results of the simulation study are very promising. The new method increases the mean coverage, creates wider confidence intervals and provides more accurate summary effect sizes compared to the current naive method.Show less
In 2011 Clarkson and Hiscock (2011) presented several regression models for flakes with different platform types used to estimate original flake mass based on platform surface area and external...Show moreIn 2011 Clarkson and Hiscock (2011) presented several regression models for flakes with different platform types used to estimate original flake mass based on platform surface area and external platform angle in order to measure reduction intensity on lithic tools. In addition to subsampling and adding external platform surface area, Clarkson and Hiscock increased the accuracy of the regression models by using a 3D laser scanner to measure platform surface area. Most previous studies multiplied platform width and thickness as an estimate of platform surface area. In this thesis, the regression models created by Clarkson and Hiscock were tested on an archaeological sample from Colmont-Ponderosa, a Middle Palaeolithic site in Limburg, the Netherlands. Instead of a 3D laser scanner, photogrammetry was used to create 3D models. It was found that Clarkson and Hiscock’s models are not applicable on the Colmont-Ponderosa sample. New models were created using the same procedure as Clarkson and Hiscock. In addition to platform type subgroups, flake shape subgroups were made. Creating subsamples based on platform type did not influence the correlation between mass and platform surface area. Subsampling based on flake shape resulted in slightly increased correlation, probably because broader flakes have a higher mass to platform surface area than other flakes. Even though a positive linear correlation between external platform angle and mass was found, this variable was not very influential on the final regression models. It was concluded that the newly created regression models are much better at predicting original flake mass for the Colmont-Ponderosa sample. Mass predictions of individual flakes are still not very accurate, which might result in faulty results when introducing new data.Show less
