# The Saber's Edge: Stars are the Riskiest Picks

Last week, I examined how actual plate appearances deviate from projections. Now, I move to the next step, how overall hitting projections deviate from the results.

While owners might think the top players are the safest options, they aren't. Stars are the riskiest. Additionally, players with perceived risk provide the most future value because they have the chance to add playing time.

I wish I could just jump to the conclusions, which are simple to understand, but I need to run some calculations to back up the results. For my analysis, I will use 2010 to 2016 Steamer projections and find the actual results. To standardize the hitter values, I used 5-category (AVG, not OBP), 12-team SDP values from the 2015 NFBC season. Using a single equation allows all the stats to be combined into pre- and post-projection values spanning several seasons.

If I have any trepidation, it is with using the same equation for all league sizes. But it shouldn't matter as I am examining how expected values deviate from projections. The single formula should have component values close enough for the years examined.

Additionally, I am going to assume no position inflation for catchers or shortstops. While it might be interesting to see how position projections differ, I will limit the research to the top projected players regardless of position.

Finally, I combined the hitters into twenty 90-player groups by their expected production to compare to their actual production.

Like last week, I found the results starting

Last week, I examined how actual plate appearances deviate from projections. Now, I move to the next step, how overall hitting projections deviate from the results.

While owners might think the top players are the safest options, they aren't. Stars are the riskiest. Additionally, players with perceived risk provide the most future value because they have the chance to add playing time.

I wish I could just jump to the conclusions, which are simple to understand, but I need to run some calculations to back up the results. For my analysis, I will use 2010 to 2016 Steamer projections and find the actual results. To standardize the hitter values, I used 5-category (AVG, not OBP), 12-team SDP values from the 2015 NFBC season. Using a single equation allows all the stats to be combined into pre- and post-projection values spanning several seasons.

If I have any trepidation, it is with using the same equation for all league sizes. But it shouldn't matter as I am examining how expected values deviate from projections. The single formula should have component values close enough for the years examined.

Additionally, I am going to assume no position inflation for catchers or shortstops. While it might be interesting to see how position projections differ, I will limit the research to the top projected players regardless of position.

Finally, I combined the hitters into twenty 90-player groups by their expected production to compare to their actual production.

Like last week, I found the results starting at the top 10 percent. Then, I move down to the 25, 50, 75 and 90 percent ranks. Additionally, I included the median number of projected and actual plate appearances. Here are the results.

 TOP PROJ. HITTERS 10% 25% 50% 75% 90% PROJ. SGP ACTUAL SGP DIFF Std Dev SGP PROJ. PA ACTUAL PA DIFF Std Dev SGP 1 - 90 3.2 0.2 -2.5 -5.5 -9.7 15.2 12.7 -2.5 5.3 673 642 -32 5.3 91 - 180 3.4 0.8 -1.3 -5.0 -8.2 13.0 11.6 -1.4 4.7 648 635 -13 4.7 181 - 270 2.9 0.5 -1.1 -4.4 -7.4 11.6 10.4 -1.2 3.8 635 625 -10 3.8 271 - 360 4.0 1.2 -1.7 -4.8 -7.9 10.7 8.8 -2.0 4.5 623 606 -17 4.5 361 - 450 4.9 1.9 -0.2 -3.6 -8.1 10.1 9.7 -0.4 4.5 613 612 -2 4.5 451 - 540 3.7 1.6 -1.9 -5.1 -7.9 9.4 7.5 -2.0 4.3 599 565 -35 4.3 541 - 630 3.9 2.6 -1.1 -5.3 -9.0 9.0 8.0 -1.1 4.9 586 579 -7 4.9 631 - 720 3.2 0.4 -1.5 -4.4 -6.8 8.6 7.0 -1.6 3.8 590 554 -36 3.8 721 - 810 3.9 2.0 -1.0 -3.0 -6.1 8.0 7.1 -0.9 3.8 575 575 0 3.8 811 - 900 3.9 1.6 -1.8 -4.4 -6.9 7.5 5.7 -1.8 4.5 561 558 -3 4.5 901 - 990 4.7 1.8 -0.4 -4.4 -7.2 7.1 6.5 -0.6 4.4 542 506 -36 4.4 991 - 1080 3.2 0.7 -1.7 -3.7 -6.1 6.7 5.1 -1.6 3.6 539 476 -63 3.6 1081 - 1170 3.0 0.6 -1.8 -5.2 -8.2 6.1 4.5 -1.6 4.4 518 459 -59 4.4 1171 - 1260 3.9 1.8 -1.2 -4.2 -6.5 5.8 4.7 -1.1 4.3 501 451 -50 4.3 1261 - 1350 4.6 1.4 -1.0 -3.2 -4.4 5.4 4.3 -1.1 3.4 483 470 -13 3.4 1351 - 1440 3.7 1.5 -1.8 -4.1 -6.2 4.9 3.1 -1.8 3.8 464 439 -25 3.8 1441 - 1530 4.2 1.7 -1.8 -4.7 -6.0 4.4 2.7 -1.7 4.1 446 388 -58 4.1 1531 - 1620 4.1 1.7 -0.8 -3.7 -5.4 3.9 3.1 -0.8 4.2 425 362 -63 4.2 1621 - 1710 3.7 1.8 -0.9 -3.7 -5.6 3.3 2.5 -0.8 3.5 398 368 -30 3.5 1711 - 1800 4.1 1.9 -0.5 -2.9 -4.9 2.8 2.2 -0.6 3.5 367 344 -23 3.5 Average 3.8 1.4 -1.3 -4.3 -6.9 7.7 6.4 -1.3 4.2 539 511 -29 4.2

The table contains a lot of data that lacks any noticeable trends. Besides the "1-90" group, almost all the group values hover around the overall averages. After analyzing the data too long, I have come to one conclusion: "The bigger they are, the harder they fall." The top players just have a larger floor because of lost playing time, mainly from injuries. They concept can be visualized with this image.

The trend shows the 10 and 25 percent breakouts are relatively linear. All projected players show the same chances of breaking out. Moving to the 75 and 90 percent performance lines, the drop-offs are larger for the higher projections.

Let this sink in: the better hitter, the riskier he is. It's not just the top handful of players providing injury risk, but the risk spread out among all the players. Mike Trout and Alcides Escobar have the same injury chances, and trying to figure out which is the bigger risk is almost impossible.

Time to move on and examine the data differently. Instead of grouping by projections, I grouped the hitters by actual minus expected results.

 HITTERS PROJ. SGP ACTUAL SGP DIFF PROJ. PA ACTUAL PA DIFF 1 - 90 7.0 14.1 7.1 533 644 111 91 - 180 7.6 12.0 4.4 546 639 93 181 - 270 7.7 10.9 3.2 558 614 56 271 - 360 7.3 9.6 2.4 555 605 50 361 - 450 6.5 8.2 1.7 540 583 43 451 - 540 7.4 8.4 1.0 560 588 28 541 - 630 7.2 7.7 0.5 558 586 28 631 - 720 7.2 7.1 -0.2 561 559 -2 721 - 810 7.4 6.7 -0.7 550 548 -2 811 - 900 6.8 5.8 -1.0 546 531 -15 901 - 990 7.9 6.4 -1.5 568 530 -38 991 - 1080 6.7 4.7 -2.0 540 497 -43 1081 - 1170 7.2 4.5 -2.7 548 481 -67 1171 - 1260 7.4 4.1 -3.3 563 453 -110 1261 - 1350 6.6 2.6 -4.0 553 398 -155 1351 - 1440 6.4 1.7 -4.7 523 320 -203 1441 - 1530 7.1 1.4 -5.7 539 334 -205 1531 - 1620 7.9 1.6 -6.4 564 303 -262 1621 - 1710 7.3 -0.6 -7.9 562 213 -349 1711 - 1800 9.3 -0.5 -9.8 592 217 -376 Average 7.3 5.8 -1.5 553 482 -71

I'll start with the top three groups. These players are desirable. They are the ones with the biggest improvements. These groups had projected plate appearances near or less than the average. They had room to play more. Also, they are productive players who achieved more playing time mainly because of season-long health or being rookie callup.

For reference, here are the top-30 players whose results were more than their projections.

 PLAYER YEAR PA R HR RBI SB AVG PROJ. SGP PA R HR RBI SB AVG ACTUAL SGP DIFF Mike Trout 2012 390 51 9 38 20 .254 5.8 639 129 30 83 49 .326 22.2 16.4 Jose Bautista 2010 410 50 13 45 3 .240 4.0 683 109 54 124 9 .260 18.3 14.4 Jonathan Villar 2016 320 32 7 30 20 .238 3.7 679 92 19 63 62 .285 17.8 14.2 A.J. Pollock 2015 523 55 9 46 15 .260 6.0 673 111 20 76 39 .315 17.9 11.9 Chris Davis 2013 456 58 23 67 2 .264 7.4 673 103 53 138 4 .286 19.0 11.6 Jean Segura 2016 507 50 7 44 20 .270 6.2 694 102 20 64 33 .319 16.2 1.0 Michael Brantley 2014 559 61 8 55 13 .272 6.8 676 94 20 97 23 .327 16.4 9.6 Curtis Granderson 2011 601 86 23 76 12 .256 1.5 691 136 41 119 25 .262 2.1 9.5 Anthony Rendon 2014 507 55 11 52 3 .264 5.0 683 111 21 83 17 .287 14.1 9.1 Matt Kemp 2011 687 85 26 105 23 .275 14.8 689 115 39 126 40 .324 23.8 9.0 Bryce Harper 2012 275 30 8 30 9 .259 2.6 597 98 22 59 18 .270 11.6 8.9 Angel Pagan 2010 344 37 6 39 14 .272 4.3 633 80 11 69 37 .290 13.1 8.8 Adam Duvall 2016 293 33 15 42 2 .244 3.1 608 85 33 103 6 .241 11.8 8.7 Chase Headley 2012 633 74 12 54 12 .266 7.6 699 95 31 115 17 .286 16.2 8.6 Matt Carpenter 2013 457 51 9 51 3 .268 4.4 717 126 11 78 3 .318 12.8 8.4 Andres Torres 2010 339 34 7 34 11 .239 2.8 570 84 16 63 26 .268 11.2 8.4 Manny Machado 2015 557 67 17 64 6 .273 7.7 713 102 35 86 20 .286 16.1 8.3 Edwin Encarnacion 2012 551 72 21 73 5 .258 8.4 644 93 42 110 13 .280 16.4 8.0 Marlon Byrd 2013 353 41 7 42 2 .254 2.7 579 75 24 88 2 .291 1.7 8.0 Josh Donaldson 2015 618 81 26 87 5 .265 1.6 711 122 41 123 6 .297 18.5 7.8 Bryce Harper 2015 560 74 23 76 9 .279 1.4 654 118 42 99 6 .330 18.1 7.8 Victor Martinez 2014 567 66 13 69 2 .289 7.6 641 87 32 103 3 .335 15.3 7.7 Daniel Murphy 2016 534 61 9 58 6 .304 7.3 582 88 25 104 5 .347 15.0 7.7 Lucas Duda 2014 376 39 12 41 2 .231 2.6 596 74 30 92 3 .253 1.3 7.7 Nolan Arenado 2015 573 69 20 76 3 .283 9.0 665 97 42 130 2 .287 16.6 7.6 Carlos Gomez 2012 370 37 7 32 19 .240 3.9 452 72 19 51 37 .260 11.5 7.6 Carlos Gonzalez 2010 620 81 23 97 19 .283 13.4 636 111 34 117 26 .336 2.9 7.5 Brian Dozier 2016 673 85 20 69 14 .241 9.2 691 104 42 99 18 .268 16.7 7.5 Alex Gordon 2011 612 71 17 64 9 .240 7.3 690 101 23 87 17 .303 14.7 7.5 Todd Frazier 2014 518 55 18 61 6 .241 6.1 660 88 29 80 20 .273 13.5 7.5

In all 30 cases, their actual plate appearances were higher than the projected amount. Additionally, they all performed better than expected, which is why some got more than expected playing time. Of the 30, 29 saw their batting average jump by an average of 33 points. Only Adam Duvall bucked the trend with a .241 AVG when it was projected at .244.

As for the players who dropped the most, it's just a list of Kyle Schwarbers. Players who played a few games but spent most of the season on the disabled list.

Now for the guts of the data. How can it be used to help our fantasy teams? I find its usefulness differs depending on if rosters are filled via draft or auction.

Drafts

The rule for using the preceding analysis for drafts is simple - when given the choice between similarly projected hitters, take the one projected for fewer plate appearances. To have equal overall values, one hitter's per-plate-appearance production has to be more than the other available options, but he has the opportunity to hit more. All the hitters have a chance to drop. An owner might as well take the player with plate appearance upside.

Auctions

I am becoming a proponent of spreading the risk around with auctions. The best players have the same injury risk as other full-time players. So a team might want to spread its money to multiple resources to limit itself from being completely out of contention if its star misses the season.

This method's key is to stay away from \$1 to \$4 players, especially in shallow leagues. Similar players are readily available and can be picked up on the cheap. Instead, I set a hard floor of \$5 per player (\$7 soft). I don't want replacement players but good players across the board. If they happen to underperform, they might still be better than replacement level guys. Additionally, I am not out of contention if my \$35-plus player goes down.

As for figuring out players to target, it's about impossible since only one player is available for bidding at a time. Two players can be compared but what are the odds an owner can pay a reasonable price for the other player?

Finally, if the auction has a reserve round, use the above draft procedure to find players with upside.

Conclusion

The idea that the best players might not be the best values contradicts the perception and actions of most fantasy owners. Owners pay a premium to avoid perceived risk by taking the best players, but they shouldn't. Most of the risk involves losing playing time. Additionally, if an owner is looking for upside, he needs to consider batters with playing time concerns. If a player is to break out from the projections, it likely will be a recently injured player or a rookie. Don't avoid these players, they should be targeted.