Bar charts are a popular and commonly used tool for the interpretation of datasets; however, representing datasets with multi-scale variation is challenging in a bar chart due to limited viewing space. To address this limitation, we introduce three novel data visualizations, called perspective charts, based on the concept of size constancy in linear perspective projection. Each of our designs focuses on the representation of datasets with important variation in the data at multiple scales. Through a user study, we measure the effectiveness of our designs for representing these datasets in comparison to traditional methods, such as a standard bar chart or a broken-axis bar chart, and state-of-the-art methods, such as a scale-stack bar chart. The evaluation reveals that our designs allow pieces of data to be visually compared at a level of accuracy similar to traditional visualizations. We also integrated our designs into a larger application for geospatial visualization of COVID-19 data. A secondary evaluation reveals that this application could be used to retrieve data more accurately than with standard bar chart visualizations. For datasets with important variation at multiple scales, the designs we present in this thesis are demonstrated---through multiple evaluations---to have advantages compared to other state-of-the-art visualizations.