In my experiment, I am doing hierarchical agglomerative clustering of texts (parameters: cosine, average). My features matrix is very sparse, so I considered PCA as dimensionality reduction technique. The first dendrogram shows the clustering of the data without PCA. The second shows the clustering of the data with PCA, but the distance on the y-axis is not in a range 0-1, even though I used cosine similarity. Can someone explain why this happened? Should I avoid to use dimensionality reduction?