Utility value of management tools Advanced Herd Management Anders - PowerPoint PPT Presentation

Utility value of management tools Advanced Herd Management Anders Ringgaard Kristensen

What kind of tools? � Complex systems � Not just single ”pieces of information” � Management information systems � ”Bedriftsløsningen” � ”E-kontrol” � Monitoring tools � Farm Watch � Decision support tools

Why do we want to evaluate? � Farmers who consider to buy a system would like to know the expected benefit � Developers who wish to sell a system would like to be able to demonstrate the benefit � Only very little research has been done in this field

Basic problems � Value of single ”pieces of information” is difficult to assess. � Secondary effects � Positive: Increased focus � Negative: Decreased focus in other areas � The farmer perhaps doesn’t use the system in an optimal way. � Interactions production system/farmer/tool � No control (what would have happened without the tool?)

Methods (Verstegen et al. 1995)

Methods (Verstegen et al. 1995) � Normative approaches � Decision theoretical approaches � Decision tree analysis � Baysian Information Economics � Control Theory � Decision analytical approaches � Simulation � Linear and dynamic programming

Methods (Verstegen et al. 1995) � Normative approaches � Decision theoretical approaches � Decision tree analysis Not value � Baysian Information Economics of tools � Control Theory � Decision analytical approaches � Simulation � Linear and dynamic programming

Methods (Verstegen et al. 1995) � Normative approaches � Decision theoretical approaches � Decision tree analysis Not value � Baysian Information Economics of tools � Control Theory � Decision analytical approaches � Simulation Examples � Linear and dynamic programming

Dynamic programming Policy 1 2 3a 3b Milk yield, kg/cow/year 7 082 6 896 7 350 6 991 Average week of replac. 25 28 21 25 Annual replacement % 50 35 59 38 Net ret., DKK/cow/year 9 236 9 150 9 544 9 319 ” , DKK/(1000 kg milk) 1 304 1 327 1 299 1 333 Number of cows 100.0 102.7 96.4 101.3 Kristensen & Thysen (1991)

Dynamic programming � Validity � What would the farmer do without the tool? � Would he/she follow the recommandations? � Are the registrations correct? � External validity: � Model versus real world � The tool tests itself � Bias for optimal policy

Simulation (Markov chain) Jalvingh et al. (1992)

Simulation (Markov chain) � Validity � What would the farmer do without the tool? � Would he/she follow the recommandations? � Are the registrations correct? � External validity: � Model versus real world � The tool tests itself � Bias for optimal policy

Simulation (Monte Carlo) Jørgensen & Kristensen (1995)

Simulation (Monte Carlo) � Validity � What would the farmer do without the tool? � Would he/she follow the recommandations? � Are the registrations correct? � External validity: � Model versus real world � The tool does not tests itself � No bias for optimal policy � The preferred normative approach

Empirical (”positive”) approaches � Verstegen et al. (1995): � Experimental designs � Field experiments � Experimental Economics � Quasi-experimental designs � Nonexperimental designs � Use of data from herds

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

Designs � No control: � Only farms using the tool are included in the study � Nonequivalent control (quasi-experimental design): � A control group is included in the study afterwards � As equal as possible to the farms using the tool � True control (experiment in the usual sense) � Farms are randomly divided into two groups � One group is told to use the tool � The other group is not allowed to use it

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

� Not serious! Time PO: Posttest only Result

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

NPO/TPO: Posttest only � Manipulation Result � Confounding between farmer type, production system and use of tool } Effect Time

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

PP: Pretest and posttest � Manipulation Result � Perhaps a general trend: All farms may have improved as those } Effect being investigated � Confounding between general development and effect of tool Time

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

NPP: Pretest and posttest Result � Correction for � General trend � Confounding with } Effect farmer type (partially, no randomization) Time

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

TPP: Pretest and posttest Result � Correction for � General trend � Confounding with } Effect farmer type (randomization) Time

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

TS: Time series, no control Result � Confounding with farmer type } Effect � Development over time � Value in the beginning versus full value Time

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

NTS: Time series, control Result � Development over Effect: b - a time } � Value in the beginning b versus full value � No confounding with a farmer type Time

Classification of designs Time Pretest- Posttest series (TS) posttest only (PO) (PP) True control (N) TTS TPP TPO Nonequivalent NTS NPP NPO Control (N) No control TS PP PO Verstegen et al. (1995)

TTS: Time series, true control Result � Development over time } Effect � Value in the beginning versus full value � No confounding with farmer type Time

Example: NTS � Value of a management information system for sow herds � Response: Piglets/sow/year � Nonequivalent control � Time series