The primary objective of this thesis is to construct a predictive model that can be used to study the Stone Age on insular, partly submerged and submerged landscapes of Greece. The chosen research...Show moreThe primary objective of this thesis is to construct a predictive model that can be used to study the Stone Age on insular, partly submerged and submerged landscapes of Greece. The chosen research area is the central Ionian Sea, as it combines diverse landscapes that have undergone dynamic changes due to eustatism and high tectonic activity. In detail, the model aimed at studying various environmental factors and their effect on the distribution of finds across the three main time periods of the Stone Age, both in terms of their original deposition and their present-day location. In addition, it aimed at studying and integrating social and cultural factors, to explore the available digital material and to use primarily open source data. The materials used were known locations of archaeological finds based on the “Prehistoric Stones of Greece” dataset, and a variety of digital maps, retrieved by European Union sites such as Copernicus and EMODnet and Natura 2000, and by national sites, for example YPEN. These datasets were open source with various Creative Commons licenses. The resolution of maps varied across each source. In order to properly examine the datasets and assess their contribution, the following process was followed. First, the known locations per time periods were split in two parts, from which the 70% was used in building the model and the rest 30% was kept for testing the model. Subsequently, the known locations were studied along with a series of maps in order to establish patterns, which were then compared to the literature. The main potential disturbances of soil and factors hindering research and findability of finds were also considered. The main factors affecting the distribution were considered to be proximity to water, elevation, depth and landslide susceptibility. No social or cultural factors could be integrated in the model. Three more factors were modelled and added, including the Natura 2000 areas, forested and increase artificial disturbances areas. Six predictive values were created, with number (1) combining low elevations (<100masl) and proximity to modern-day water bodies, and the lowest (6) being the underwater areas with depth higher than -1000. One model was created for all three time periods, due to the overall similarity of observed patterns. The resulting model was tested with the withheld sample of locations and it showed that the values carrying the majority of finds are Value 3 and 1 for the Palaeolithic and the Mesolithic, and Value 1 and then 3 for the Neolithic. In conclusion, post depositional processes seem to have largely affected the 150 distribution, but predictive modelling can still be effective. In term of social and cultural factors, more research is needed before they can be integrated in a model, especially on the first two periods of the Stone Age. Finally, it is possible to create a predictive model of the Greek Stone Age by using mostly open source material and open data.Show less
Psychological research is often based on null hypothesis significance testing (NHST) using explanatory modelling. In many cases, additional information or even more reliable information can be...Show morePsychological research is often based on null hypothesis significance testing (NHST) using explanatory modelling. In many cases, additional information or even more reliable information can be gained if predictive modelling is also used. Cross-validation (CV) is a very useful statistical procedure that can generate predictive outcome measures. This study will compare the use and results of CV with the results of traditional NHST analyses using two case studies, one explanatory theory-driven - comparing means - question and one predictive data-driven - forwards stepwise logistic regression - question. In the case of explanatory questions, CV is able to generate similar conclusions and the outcome measure is more intuitive to interpret. Regarding the predictive data-driven question, the final models from the CV procedure have slightly lower out-of-sample prediction errors than the final model based on the traditional NHST procedure. Moreover, CV proves useful in evaluating explanatory formulated models with respect to out-of-sample prediction accuracy and overfitting. It is recommended to implement predictive modelling in contemporary psychological research complementary to explanatory modelling in order to make psychological science more reliable and replicable.Show less
