Component 1: explains the difference between Viena variants.
Component 2: explains the difference between Pavo variants.
Notable variables: Floury, Rubbery, Dark Meat, White Meat, Umami (aka HVP), Spicy, Bitter, and Acidic.
We used PCA to analyze both the Observations and Variables of our dataset. We chose not to scale or center by row by Panelist because we assumed that the Panelists’ ratings are all on a same scale. The results yielded from the Observation Factor Score and Variable Factor Score (Loadings) - colored by Panelists also confirmed the idea that all Panelists rated relatively similarly. The most noticeable results is the spread of data points along component 1 and 2 according to their sausage type (Pavo vs Viena). We can see clearly that Component 1 explains the difference between Viena (mixed meat) variants while Component 2 explains the difference between Pavo (Turkey) variants. Based on the narrative of Viena vs Pavo, we can interpret from the contribution barplots (verified by bootstrap resampling) that important characteristics include Floury, Rubbery, Dark Meat, White Meat, Umami (aka HVP), Spicy, Bitter, and Acidic. We can see several pattern / grouping / theme among these important characteristics like “textures” vs “meat types”, “basic taste” vs “others”, “bitter” vs “others”, etc. We will explore these groupings in variables with other methods in the following chapters.
Note on drawbacks of PCA: PCA is only effective to detect linear relationships. Thus, PCA can leave out a lot of information in datasets where the observations or variables have non-linear relationships. Another problem of PCA is that it can only analyze quantitative data (as opposed to qualitative - MCA, CA)