EK Peerages, a slight analysis
Apparently due to a conversation on the EK general discussion list (which I am not on), scaharp posted some counts of elevations-per-half-decade in the East kingdom for each of the peerages. I grok graphs and stats some, and I wanted to break it down by years.
This is pretty much a copy-and-paste from a comment I made in her livejournal:
Let's look at the raw-numbers: This graph is number-of-new-elevations-per-year for the three peerages:
Kinda hard to follow, but you can see that the trend right now is generally downward - some of that is because I included 2011 in the data, and we haven't finished.
First, I did a 10% smoothing - this is the way speedometers don't jump all over the place constantly: "For the first number, use that number. For the second, take 90% of the previous smoothed value and add 10% of the new data point". It's great for finding trends, but it's not really the right kind of analysis for this data set, it's meant when the previous data doesn't change much. It's great for tracking your actual weight, if you're dieting.
Last, I did something more relevant to the effectively-independent numbers: weighted average of five points. 10% of year X-2, 20% of year X-1, 40% of this year, 20% of next year, 10% of year X+2 - obviously, only doable until two years ago. This is the kind of smoothing you use when the numbers are related, but there can be a lot of change between data - weather forecasters use this so that a single day of "no history of storms" in the middle of storm season doesn't cause a bad forecast.
Sure, maybe only 20% of the SCA fights, but that doesn't seem to have been relevant to the number of elevations relative to the other two peerages until around 1990.
This is pretty much a copy-and-paste from a comment I made in her livejournal:
Let's look at the raw-numbers: This graph is number-of-new-elevations-per-year for the three peerages:
Kinda hard to follow, but you can see that the trend right now is generally downward - some of that is because I included 2011 in the data, and we haven't finished.
First, I did a 10% smoothing - this is the way speedometers don't jump all over the place constantly: "For the first number, use that number. For the second, take 90% of the previous smoothed value and add 10% of the new data point". It's great for finding trends, but it's not really the right kind of analysis for this data set, it's meant when the previous data doesn't change much. It's great for tracking your actual weight, if you're dieting.
Last, I did something more relevant to the effectively-independent numbers: weighted average of five points. 10% of year X-2, 20% of year X-1, 40% of this year, 20% of next year, 10% of year X+2 - obviously, only doable until two years ago. This is the kind of smoothing you use when the numbers are related, but there can be a lot of change between data - weather forecasters use this so that a single day of "no history of storms" in the middle of storm season doesn't cause a bad forecast.
Sure, maybe only 20% of the SCA fights, but that doesn't seem to have been relevant to the number of elevations relative to the other two peerages until around 1990.