Given a set cache size with 1-way associativity, why would increasing the block size decrease the miss rate? With a fixed cache size, don't we store the same amount of data, meaning we would obtain our data just as accurately?
Good question.
With a fixed cache size, increasing the block size would cause us to store less blocks, keeping the net size of data the same, but the data stored (in the cache) would be less spread out in memory overall, since the data belong to fewer blocks total overall. So, due to the principle of spatial locality, we would have to reach out into memory few times to get new blocks. (This is because we would use the memory we would already have more likely than we are to otherwise.) Hence, a decrease in the miss rate.
However, the miss penalty would be increased due to this. This is because in the event of a miss, we would need to copy more data into the cache. (We would have to load a 16 KB block into the cache instead of a 4 KB block.