How to Conduct On-Farm Trials

HowtoConductOn-FarmTrials

Dr.JimWalworthDept.ofSoil,Water&EnvironmentalSci.

UniversityofArizona

Howcanyoudeterminewhetheratreatment(thismightbeanadditive,afertilizer,snakeoil,etc.),cultivar,ormanagementpracticeiseffective?

Testitforyourselfinyourorchard• Treatpart(s)ofyourorchardwiththematerial• Leavepart(s)untreated• Measuretreeresponse

– Yield– Leafanalysis– Nutquality– Etc.

• Compare responseoftreatedversusuntreated

Howdowedecideifsomethingisbetter?

• Weneedatleasttwo‘treatments’sowecanmakeacomparison

• Weneedanobjectivewayofevaluatingthetwotreatments• Statisticscanhelpusdetermineifatreatmentactually

affectsplantperformance

Onaveragewho’staller- menorwomen?

Men’sheights Women’sheights

5’6” 6’0” 6’2” 5’11” 5’10” 5’6” 5’7” 5’2” 5’0” 5’6”

66” 72” 74” 71” 70” 66” 67” 62” 60” 66”

Inoursamplesthereisarangeofheightsofbothmenandwomen.Men’sheightsrangefrom66to74”,women60to67”. Somewomenaretallerthansomemen.

Samplestatistics

• Averagesormeans

men women66” 66”72” 67”74” 62”71” 60”70” 66”

averages 70.6” 64.2”

5’10” 5’4”

( ) 2.645

6660626766=

++++

Wecantaketheaverageofeachgroup,butwiththevariabilityinmen’sheightsandinwomen’sheights,howdoweknowifthesenumbersreallydifferent?

• Westartbyevaluatingthespread(statisticiansusemeasuresofvariance orstandarddeviation)

Freq

uency

Value(women’sheight)

5’1”5’2”5’3”5’4”5’5”5’6”5’7”

Mean(average)

StandardDeviation

coefficientofvariation = standarddeviation

mean ×100

standarddeviation = variance�

Groupsorpopulationswithlarge(green)andsmall(red)spreads

Comparinggroupswithsmallversuslargespreads

64.2 70.6 64.2 70.6women men women men

AnalysisofVariance(ANOVA)usingMicrosoftExcel

InMicrosoftExcel,enterthedataintwocolumns,oneforeach‘treatment’(heremenandwomen)

InMicrosoftExcel(OfficeProfessional2016),goto‘File’

Thento‘Options’

Thento‘add-ins’

SelectExcelAdd-insandclick‘Go’

Select‘AnalysisToolPack’and‘OK’

Nowwhenyouclickon‘Data’you’llseea‘DataAnalysis’tabovertotheright.

Select‘Anova:singlefactor’

Select‘Anova:singlefactor’

TellExcelwheretofindyourdataandwhereyouwanttheanalysisoutput

Averages

ThePvaluetellswhetherornotthetwomeansaresignificantlydifferent.ThePvalueistheprobabilitythatthetwomeansarenot different.Inthiscasethereisa0.97%chance

thatthemeansarenotdifferent,ora99.03%chancethattheyreallyaredifferent.

Conclusion:Menaretallerthanwomen!

Note:inscientificresearchwegenerallyconsidermeanstobedifferentiftheANOVAtellsusthereisatleasta90-95%chancethattheyaredifferent.

Orchardresearch• Youneedatleasttwotreatmentssoyoucanmakea

comparison• Treatmentsmustbereplicated (weliketoseeatleast4

replicates)– Thedifferencebetweenreplicatesofeachtreatmentgiveusameasure

ofthevariance(spreadorvariabilityofthedata)– Replicatesshouldbearrangedinarandomizedfashionsothat

differenceinthefield(soilproperties,slope,elevation,water,etc.)donotunfairlyinfluenceonetreatment

We’vedividedthisorchardblockintotensections,eachoneanirrigationpanel.

Let’ssupposewewanttocomparetwotreatments:AandB.Thesemightbetwofertilizermaterials.Eachpairofpanelsisonereplicate.Werandomlyassigneachpanelinareplicateoneofthetreatments,AorB.

A B A B B A A B B A

Snak

e oi

l

Snak

e oi

l

Snak

e oi

l

Snak

e oi

l

Snak

e oi

l

No

snak

e oi

l

No

snak

e oi

l

No

snak

e oi

l

No

snak

e oi

l

No

snak

e oi

l

Wecollectdatafromeachpanelindividually,sowehavefive‘A’measurementsandfive‘B’measurements.Measurementscouldbenutyield(shownonfigure),leafnutrientlevels,insectpopulations,etc.

A B A B B A A B B A3375 3455 3020 3220 3140 3005 3065 3220 3010 2925

Snak

e oi

l

Snak

e oi

l

Snak

e oi

l

Snak

e oi

l

Snak

e oi

l

No

snak

e oi

l

No

snak

e oi

l

No

snak

e oi

l

No

snak

e oi

l

No

snak

e oi

l

Accordingtothetreatmentaveragestreatment‘B’wasbetterthantreatment‘A’

Buttheanalysisofvarianceindicatesthatthereisa25%chancethattheaveragesarenotsignificantlydifferent,sowecannotsaythattreatment‘B’wasreallybetter.Thereisagoodchancethat

differencesbetweentheaverageswerejustcausedbyrandomvariability.

Inthiscase,itlookslikethesnakeoil

decreased nutyield

Otherconsiderations• Sometimesrealrandomizationisnotpossible

– Youcanstillreplicateandconductananalysisofvariance,butthere’sachancethatobserveddifferencewillbefromfieldvariation,notthetreatments

– Replicatesalwaysmustbesampledorharvestedseparately

• Itisagoodideatoleave‘buffer’rowsthatseparatetreatments,soadjacenttreatmentsarenotaffected– Thisisparticularlyimportantwiththingslikespraytreatments,but

evenhedgingonerowwillaffectthenextrow– Treatmentscouldbeappliedtoeveryotherrow

• LocalCooperativeExtensionpersonnelcanhelpyousetupandanalyzefieldresearch