The Extension Challenge of Crop Variety Selection in Colorado. LSD Alpha Level Revisited. Painting a Picture

Size: px
Start display at page:

Download "The Extension Challenge of Crop Variety Selection in Colorado. LSD Alpha Level Revisited. Painting a Picture"

Transcription

1 The Extension Challenge of Crop Variety Selection in Colorado Plus LSD Alpha Level Revisited Plus Painting a Picture Jerry J. Johnson Professor and Extension Specialist Colorado State University Crops Testing Program

2 Upon learning I am a professor at CSU I am asked: What do you do? What do you Teach? For years I tried to explain: I m in charge of Crops Testing. When there are new crop varieties I conduct field experiments to see if they are adapted to our conditions. They give me a blank stare and say--- How do you stand on GMOs? You must believe in them Then I give them a blank stare not knowing where to start--- Now, I answer them differently: Imagine you are a farmer and you have to decide which variety to plant. I give farmers unbiased and research-based data to decide.

3 Every Year, Every farmer must decide, Every field which variety to plant.

4 For the past 11,000 years: every farmer

5 Every year, Every farmer must decide

6 Every year, Every farmer, Every field

7 And, every future farmer!

8 What is the total variety decision worth for all crops tested by Crops Testing in Colorado? (Assume 2% genetic gain per year) Corn $12,828,000 Wheat $6,004,000 Sorghum $2,000,000 Proso millet $460,000 Dry Beans $453,000 Sunflower $346,000 Total potential value of variety decision annually = $22,091,000

9 The Extension Challenge for Crop Production in Colorado Approximate Cropland Area: 10.6 Million Acres Number of Farms and Ranches (2017): 33,800

10 Why I use LSD: α = 0.30 Short answer: (Other smarter people) Carmer, S.G Optimal significance levels for application of least significant difference in crop performance. Crop Sci. 16: Johnson, J.J The advantages and disadvantages of replications in onfarm sorghum variety tests in North Cameron. MS Thesis. Washington State University Johnson, J.J., J.R. Alldredge, S.E. Ullrich, and O. Dangi Replacement of replications with additional locations for grain sorghum cultivar evaluation. Crop Sci. 32: Campbell, K.G., Y.M. Thompson, S.O. Guy, M. McIntosh, and B. Glaz Is, or is not, the two great ends of Fate : Errors in Agronomic Research. Agronomy Journal 107, Issue 2: Bowman, D., D.Dombek, and J. Still Journal of Crop Variety Testing. Vol. 1:1-7.

11 REPORTING DATA IN CROP PERFORMANCE TRIALS IN THE UNITED STATES Daryl T. Bowman, Don Dombek, Joshua Still How data are reported is critical in how that data are used and interpreted This review paper looks at various aspects of data reporting using surveys and scientific articles on the subject Surveys covered wheat (Triticum aestivum L.), corn (Zea mays L.), and soybeans (Glycine max (L.) Merr.) We recommend reporting a minimum of mean and Least Significant Difference (LSD) choosing a significance level preferably between 0.20 and Table 1. Statistics reported from crop performance trials in 30 states Mean LSD. 5% LSD 10% LSD 20% LSD 25% Published in Journal of Crop Variety Testing Volume 1:1-7 (2017)

12 Why I use LSD: α = 0.30 It is important for each researcher to think about the risk structure. Is Type I error important? Is Type II error important? Type I Type II As risk of Type I error decreases: Risk of Type II error increases

13 Should I plant Variety A or Variety B that are the same for all characteristics except yield? Risk Structure for Variety Selection

14 Johnson says the data says: Variety A is better than Variety B when, in reality and unbeknownst to Johnson, the two varieties are equal yielding. Whichever variety I choose, I don t lose anything. Johnson just loses face because he committed Type I error, I suffer no economic loss. I don t care about protecting myself against Type I error

15 Johnson says that the data says: Variety A is equal yielding as Variety B when, in reality and unbeknownst to Johnson, the two varieties are not equal yielding. Half of the time I will choose the variety that is actually lower yielding, I suffer loss half of the time Johnson doesn t just lose face because he committed Type II error, I suffer economic loss. I care about protecting myself against Type II error

16 Carmer s Argument: Unfortunately Type II error risk can not be set to obtain a corresponding LSD; like for Type I error, i.e. LSD 0.05 The most practical way to control Type II error is to set a different level of Type I error. Thus α = 0.30 instead of α = 0.05

17 Every researcher must think about the risk structure. Is Type I error important? Is Type II error important? Is Type III error important? As risk of Type I error decreases Risk of Type II error increases Risk of Type II error also decreases LSD Type I ( ) = 0.30 Type I Type II Type III Type I =.001 Type II =.999 Type III =.001

18 Johnson says that the data says, Variety A is superior yielding to Variety B when, in reality and unbeknownst to Johnson, Variety A is significantly lower yielding than Variety B. Johnson doesn t just lose face because he committed Type III error, I suffer economic loss. 100% of the farmers selecting Variety A would lose yield. I care about protecting myself against Type III error

19 Implied Type III error comprised of two errors: ½ of Type II error and pure Type III (choosing to plant variety that is significantly lower yielding than the one that stats said was significantly higher yielding). A farmer could reasonably want to keep Type III error below 0.10

20

21 Breeders Capable of Breeding Many More Types of Wheat than Before 1995: CO wheat breeding program bred only Hard Red Winter Wheat Since: Russian Wheat Aphid resistance Hard White Winter Wheat High-quality wheats Clearfield wheat (Single-gene & double-gene) Disease resistance Wheat stem sawfly resistance ACCase herbicide resistance

22

23 Risk Structure for Variety Selection Is Variety B with some desirable trait equal yielding to Variety A (without the trait) but the same for all other characteristics except yield? Protect against falsely asserting Variety B (with the trait) is equal yielding as Variety A (no trait). Is there a yield drag associated with a trait? Type I error right? Different risk structure.

24 The Heart of Change J.P. Kotter & D. Cohen People change what they do less because they are given analysis that shifts their thinking than because they are shown a truth that influences their feelings

25 Difference in yield? Brand-Hybrid Yield Moisture Bu/Ac % Dekalb DKS ** 15.3 a Dyna-Gro 772B 102.8* 14.8 a Terral REV * 14.9 a b Terral REV * 15.1 a b Dyna-Gro GX * 15.0 a b Terral REV * 14.9 a b Dyna-Gro GX * 14.8 a b Dekalb DKS * 15.1 a b Alta AG * 15.1 a b Dyna-Gro M60GB * 14.8 a b Alta AG * 14.6 a b Alta AG * 14.8 a b Alta AG * 14.9 a b Alta AG * 14.5 a b Dekalb DKS * 15.0 a b Dekalb DKS b Alta AG Dyna-Gro GX Alta AG MEAN LSD (10%) CV (%) **Highest yielding variety in test *Yield not significantly less than the highest yielding variety in the test Only significant difference

26

27

28 Dryland Hard White Hard Red Premium No Premium Snowmass YR: 6 WSMV: 3 Antero YR: 2 WSMV: 5 Clearfield Non-Clearfield Sunshine YR: 5 WSMV: 8 Joe YR: 1 WSMV: 3 High Yield Single-gene Doublegene Early Maturity Medium Maturity Medium-Late Maturity High Yield Brawl CL Plus YR: 5 WSMV: 7 Langin YR: 3 WSMV: 6 Avery YR: 6 WSMV: 3 SY Monument YR: 2 WSMV: 8 Oakley CL YR: 1 WSMV: 2 YR=Stripe rust rating (1=resistant, 9=susceptible) WSMV=Wheat streak mosaic virus rating (1=resistant, 9=susceptible) WB-Grainfield YR: 2 WSMV: 8 Byrd YR: 7 WSMV: 2 LCS Mint YR: 4 WSMV: 5 Denali YR: 7 WSMV: 3

29

30

31

32

33

34

35

36