@Henry the Hedgehog: Early PC cost histogram findings
A few weeks ago you asked me about average PC cost and I replied that I thought a histogram would be more interesting. I added PC cost to the history pages this morning:
and I plan on setting up a graph page that'll automatically update a histogram of that data. So far, though, I'm finding this:
- Through 67 weeks of FML play, 57 times the cost of PC was $950 or more. That's 85% and would appear to be true this week as well.
- Average is 970
- Std Dev is 75
- 32 times, PC cost exactly $1000
I believe when I looked through this that almost every time the PC cost less than $950 it was a 7x or 8x bonus playSep 3, 2016, 11:29am PDTEdited
I stand corrected, I misread the online histogram I generated. It's 60 out of 67 weeks where the cost of PC was $950 or more, 89.6%.
To Ari's point, the 7 that aren't are indeed either x8 or x7 plays. They are:
2015 Summer Week 1 (Pitch Perfect 2 x8)
2015 Summer Week 15 (War Room x8)
2015 Fall Week 1 (A Walk in the Woods x8)
2015 Fall Week 3 (Black Mass + War Room x7)
2015 Fall Week 10 (Peanuts + The Martian x7)
2015 Awards Week 6 (Revenant + The Forest x7)
2016 Spring Week 2 (Deadpool x8)
That's some good guidance for PC chasers. You'd better spend $950 unless you think there's a x7 or x8 play.
I'll do some similar analysis on BP's when I get a chance and make both a permanent feature of analyzer.fmlnerd.com here in the coming weeks.Sep 3, 2016, 12:53pm PDT
That looks right to me! I remember that statistic because I pointed it out when you were talking about recalibrating your model (is that ever happening btw?) and I suggested you only consider lineups that cost $950+ and every possible combination with a 7x and 8x playSep 3, 2016, 1:39pm PDT
@Ari - Reworking the model is still in plan, but a couple of things happened there. I only have so many hours a week to devote to FML projects and what I quickly realized in the Spring when you and I had that back and forth is that I didn't have good enough data to get much more sophisticated than what I already had. I needed a way to capture data that I could then work through programmatically so I could run regressions properly to test new theories.
Once I figured out how to do that and I could automate most of my weekly activities (including World Cup/Olympic scoring), it spawned ideas for all these other tools, which I'm slowly releasing now through the new site. At some point I'll be far enough along with that so that I can revisit the model with a lot more power/data behind it. Until then, the current model is enough of a conversation starter that I find it gives me enough to write about each week, I just take a more skeptical eye to its output than I once did.
BTW, for those interested I have in flight:
Sep 3, 2016, 3:04pm PDT
- Thursday preview data interactive scatterplot (courtesy my intern)
- Histograms for PC cost, BP cost, and screen combinations
- A prettier Friday report (probably by next week)
- Pick rate graphs that show how average/percentage screens change as the week goes on (the week after that)
OK for BPs it's a little different distribution.
- 41 of 67 (61.2%) are between 7 and 106
- 50 of 67 (74.6%) are between 7 and 207
- Only 7 of 67 (10.4%) are above 507
- Average 183.5
- Std Dev 232.5
Again, I'll figure out a way to pretty this up and automate it here in the coming weeks, but there appears to be more variance for BP than for PC.Sep 3, 2016, 4:08pm PDT
What about the wisdom of crowds?
More data for thought, this time the delta between the PC score and the Most Popular score.
Sep 3, 2016, 6:14pm PDT
- Out of 67 weeks, Most Popular has also been PC 11 times (21%)
- The biggest delta between the two was Summer 2015 Week 1 (Picture Perfect x8): $ $143,200,190. Clearly people were still figuring out how to play.
- The next biggest delta was Summer 2015 Week 9 (Trainrwreck x6, Inside Out, Ted2): $96,183,938
- Average delta: $21.7M
- Standard Deviation: $25.2M
- 70% of the time, the delta was below $25M
This is great data and certainly gives us a great idea of how to spend our bux. Always looking for ways to minimize risk, and good data helps a lot! Thanks Nerd for your awesomeness!Sep 3, 2016, 8:07pm PDT
One more set of findings and I'm done for the day. The number of different movies per PC break down like this:
- 1: 4 times
- 2: 19 times
- 3: 24 times
- 4: 17 times
- 5: 3 times
So, 60 out of 67 times (89.5%) PC consisted of 2, 3, or 4 different movies.
7x screens of a Best Performer combined with one other screen has won PC 11 times, the most often occurring two movie combination (as opposed to, say 6x screens and 2x screens, which has happened 4 times).
Similarly 6x screens of a Best Performer combined with two other single screens has won PC 8 times, the most often occurring movie combination for using only three movies.
The most often occurring four movie combination is 5x Best Performer used with three other single screens, which has won PC 7 times.
What about blank screens? 65 out of 67 times (97%), PC did not have a blank screen. No PC has had more than one blank screen in the first 67 weeks of Fantasy Movie League.Sep 3, 2016, 9:08pm PDT