Ecological Analytics has provided statistical services and solutions for academics and Government research institutes. Additionally, we have supported medical, commercial and economic sector projects.
Ecological Analytics uses open source software for all analyses. We generally use R for traditional statistical analyses, and Python for Machine-Learning approaches. Graphics can be supplied in R (lattice, GGPLOT) or Python (Matplotlib, Seaborn) as required. We favor Jupyter Notebooks for both platforms to enable clients to easily understand and modify code.
Regression forms the cornerstone of many exploratory data analyses. Coupled with regression trees, linear and non-linear regression can be applied to dissect data sets and gain the best insights from the data available. Data can be costly to obtain. At Ecological Analytics we believe in interacting with the data to derive the most we can while ensuring the results are robust.
Generalized mixed effects analyses can deal with models ranging from logistic, through normal to skewed distributions such as lognormal and negative binomial. Overdispersion, or the presence of more zeroes than expected, can be accounted for in quasi models, hurdle or zeroinflated models as required. If the model contains random factors, we will ensure these are correctly specified in the model.
When the project objectives require analytical prediction, Machine Learning techniques come into play. We have experience and expertise in a wide range of methods ranging from Support Vector Machines, Random Forests, and Gradient Boosting machines through to Recurrent Neural Networks including LTSM models for time series.
One study objective may be to identify groups in data. A range of unsupervised classification approaches are available including hierarchical cluster analysis and partitioning methods such as k-means and k-nearest neighbors. Out philosophy is to apply a range of methods and tune them to find the best consensus.
Our expertise in multivariate analysis includes the full range of constrained and unconstrained ordinations, and Canonical and partial ordinations. We also can provide hierarchical and non-hierarchical cluster analyses including k-means and k-nearest neighbors. If required, we can conduct formal hypothesis testing using Multivariate Analysis of Variance, Permutation MANOVA (PERMANOVA) and randomization
Repeated measures and temporal data present analytical challenges for sampling and experimental data. Similarly, traditional time series data require a dedicated approach for description and forecasting. Ecological Analytics has experience and expertise in a range of techniques for dealing with autocorrelated data and time series description and forecasting.
Data collected over a spatial region may require particular analytical techniques. In an experimental setting, spatial autocorrelation may need to be corrected for. If the objective is to map the variable of interest, the spatial pattern may itself be important. Spatial statistical modeling such as kriging may be necessary. We have an extensive toolkit of spatial approaches.