Home page

Decision analysis as a basis for more effective agricultural innovation

In the recent bulletin covering the joint extension service offered by Navatec and Strides, Hector McNeill1, referred to the term decision analysis. The article below is Part 1 of a series of 4 with the following titles:

  • Part 1: What is Decision Analysis and how does it work?
  • Part 2: Designing resilient projects
  • Part 3: Tacit knowledge & performance
  • Part 4: The cloud-based decision analysis tool box

In this article Hector McNeill explains what decision analysis is and how it works. There are a significant factors that go into improving the application of decision analysis to agricultural development and agricultural innovation projects, including the considerations for an effective cloud operation. These issues will be covered in later articles in this series.

Part 1: What is Decision Analysis and how does it work?
The power of the keyboard


Before committing substantial public or private finance to actions it is preferable to determine the best options from the standpoint of feasibility, costs and risks through simulations, based on decision analysis models. By investing in decision analysis, millions if not billions can be saved by avoiding losses resulting from costly mistakes, lack of optimization, lack of appreciation of the dependency of a project on specific conditions and "over-runs".


Summary

Decision analysis is a precise method to identify projects, support their design and understanding of their degrees of resilience and sustainability as well as support decision-making in the face of change during implementation. It has a major potential contribution to reducing exposure to risk and increasing the returns on investment to lender and donor funding of agricultural development and innovation projects.

However, the procedures and methods used seldom feature as part of the training of agronomists, economists and those working in the scientific disciplines in the agricultural domain. A lot needs to be done on the side of training.

One solution that has been implemented in the Navatec System, the cloud software-as-a-service project cycle and portfolio management system. The foundation of this system in the decision analysis approach. Therefore by inputting data on gaps and needs and constraints, feasible solutions and resource requirements in a logical and sequential fashion, ends up with appropriate project models that can then be run as analytical simulations. Thus, by inputting design information the system handles the rest.

Training still remains an important requirement but by learning to use the Navatec System it is possible to design optimized projects and secure an ability to make effective decisions in response to change during project implementation.

Origins

The discipline of Decision Analysis was developed by Professor Ronald Howard and others who started the Decision Analysis Group in 1965 at the Stanford Research Institute in Menlo Park, California. Howard coined the term "decision analysis". Decision analysis works on the basis of a decision being defined as an irrevocable allocation of resources to specific action/s designed to achieve a specific objective. If a decision is later altered further additional resources will need to be allocated. Therefore decisions are important. Decision analysis identifies the appropriate actions based on analysis of the best available knowledge and information.

When should Decision Analysis be used?

Decision analysis is applied in situations when the required decisions are characterized by:

  • Uniqueness - each is one of a kind, perhaps similar to - but never identical with - previous situations
  • Importance - a significant portion of the organization's resources is in question
  • Uncertainty - many of the key factors that must be taken into account are imperfectly known
  • Long run implications - the enterprise will be forced to live with the results of the situation for many years, perhaps even beyond the lifetimes of the individuals involved
  • Complex preferences - the task of incorporating the decision-maker's preferences about time and risk assumes great importance

Decision analysis provides a logical framework to balance these considerations. Use is made of logic and mathematical modeling of the decision parameters in the form of a decision analysis model in a computer to simulate and evaluate the environment and various courses of action. This has been applied successfully by large organizations over the last 60 years. Today the estimated application of decision analysis involving significant savings is applied to agricultural commercial production planning, feed mixing, logistics, forecasting and primary industries, manufacturing, commodities and transport. Out of the estimated $2 trillion in global agricultural production (this is the first 52 commodities - FAO-Stat) it is estimated that around 25% or $500 billion handled by large private operations who use decision analysis yielding operational savings of around $100 billion each year. The annual international and private funding of aid, according to the OECD is around $165 billion and yet very few projects the funded projects benefit from such analysis. International economic development agencies and banks deploy conventional project cycle management procedures do not use decision analysis. This leads to lower quality projects and investment performance. Together with other factors, this gap in due diligence contributes to the exceedingly high level of project failures reported to exceed 40% of the total. This is equivalent to around $35 billion wasted resources each year.

Decision analysis has a lot to contribute to the resolution of decision issues surrounding both the understanding of environments, ecosystem interactions and sustainable production systems. This potential contribution can improve the use of resources in individual sustainable projects as well as actions and policies to be carried out at local, regional and the macroeconomic levels. The scale of operation depends upon the particular focus of a donor, investor, community or government.

Building a model

The first step in decision analysis is to identify the decision environment and the explicit relationships between the critical factors that determine the potential outcomes of different types of decision. This information is used to build a model. Since the 1960s these models have been based on input-output data such as the relationship between fertilizer input, the crop growth process and final yield of a crop. The construction of the model and use of the model involves a decision analysis cycle used to refine the model on the basis of basis of a incremental improvements in the quality of three specific types of information :
  • The information and knowledge used toOperational coherence is the degree to which construct a model which encapsulates the understanding of the cause and effect relationships between the critical factors which determine decision outcomes (deterministic)
  • The confidence in the applicability of the model rests upon the understanding or estimation of the probabilities of critical events, not all of them under the control of the decision-maker, which can affect decision outcomes (probabilistic)
  • The utility of the model finally depends upon the availability and quality of the required information (data), knowledge and understanding of determinants as well as basis for estimating the likelihood of decision outcomes (informational)
A model is deemed to be satisfactory when the marginal costs of collecting additional information are not outweighed by the apparent gains in terms of the significance of the marginal improvements in output estimates and if the model has demonstrated its ability to reproduce known input-output relationships accurately.

Detailed decision analysis

Once a model is deemed to be adequate as a basis for decision analysis it is used as a functional structure to support several types of analysis to determine:

  1. What specific combinations of resources can result in optimized processes to generate outputs on the basis of a maximization or minimization function with relation to output quantity, output quality, total and unit costs, income, margins and timing.
  2. What the effects of different feasible ranges of input values on outputs; which inputs have most impact on output.
  3. What are the effects of ranges of likely external factors that can impact a project's ability to deliver the desired output.
  4. What is the most sustainable, resilient combination of resources and process methods that can raise the likelihood of achieving the project objectives

The most frequently applied analyses each have specific types of objective and they include:

  • Linear programming
  • Markov models
  • Monte Carlo simulation (MCS)

Linear programming is used to identify the best combination of inputs to achieve a specific objective such as maximization, minimization of an objective such as output, margins or time. This identifies the quantities of each variable input required to satisfy the objective.

Markov models carry out analyses that show how both inputs and outputs will vary over time according to a model that contains the algorithms that establish the conversion of inputs into next state outputs. This can be useful, for example, in the case of population dynamics or innoculum potential dynamics in the case of animal and crop disease epidemics.

Monte Carlo simulation is used to assess the range of outcomes associated with ranges of inputs to the model by changing the inputs of interest from discrete (single) values to inputs of a likely, minimum and maximum values and a stochastic distribution of the data over this range. The simulation is a reiterative calculation of model outputs across the stochastic values for any number of inputs or outputs and to present the outcomes in a transparent manner in tabular and graphic forms.


Assessing soil fertility; the importance of soil analysis and its interpretation, Johnny Johnston, Rothamsted Research, Harpenden, AL5 2JQ, UK

Examples in agriculture

Agricultural models2 emulate existing input-output relationships as algorithms in a computer program. For example, the diagram on the left shows the variation in yield of a crop according to fertilizer inputs measured each year for 5 years. The result is three production response curves that combine years with similar responses due to differences in seasonal conditions. These conditions relate specifically to temperature and availability of water.

In this case the data comes from an experimental station but is could also have been collected by extension services or farm survey systems such as the Farm Accountancy Data Network (FADN) applied across the 28 member states of the European Union. In this example, data for 1988 shows good yields. This was the result of warm conditions, adequate water and dry harvest conditions. In 1989, there was a late cold spring followed by drier conditions leading to a lower yield. The "average" yield occurred in years 1986, 1987 and 1990 with intermediate water and temperature conditions. The important point to note is that the variation in response levels to "external "environmental factors (temperature and water regimes) was greater than the response to fertilizer within any particular year. This is an indication of the dimension of the significance of uncertainty in the case of crop production.

McNeill, H.W., "Biomass production according to EWT complex item values", 2000, SEEL.

Based on: McNeill, H.W., 3D Development Model, TP, Food Research Institute, Stanford University, 1968 and McNeill & Jino, "Simulation of 3D Development Model", CNAE, 1969.

In this simple graph there are two forces at work. One is the natural "external" environmental conditions affecting temperature and water regimes. The other is the input of fertilizer that determines the soil fertility level. This 2-dimensional graph can be reconfigured into a more useful analytical 3-dimensional production surface that make temperature and water status specific measured inputs on a temperature (T) and a water (W) axes, as shown on the right. The fertility regime is the edaphic axis (E).

As can be appreciated, this model has changed from being a crop response curve facing uncertainty due to "external" conditions into an environmental and ecosystem model in which yield or biomass is related in a predictable functional sense to the natural inputs of water and temperature with variable inputs of fertilizer, for example on a farm. The natural parameters can be estimated using normal ranges for water availability in the soil (measured as water deficit) and temperatures that exist in agro-ecological records or climatic data.

The production surface now has two important analytical functions:
  • by tracking water and temperature regimes duping the year it is possible for forecast yields (biomass)
  • any yield can be associated or explained by the cumulative average temperature and water deficit days
It is then possible to add in complementary component factors to the model such as irrigation or even changing to covered production. These will, in turn, will result in a changes response surface for the crop.

Alternative production scenarios include selecting different genotypes that are more adaptive to the prevailing environmental conditions, combined with crop rotation so as to lower costs of production and augment production resilience. We might wish to add, into the model, the unit prices/ha. of fertilizer and irrigation and also add in the cost of operations so as to calculate the gross margin on a crop

Where
Gross Margin/ha = Yield/ha. x Unit price - Variable input costs


Why is this interesting? Well the unit costs of output will vary with yield resulting in a family of possible gross margins, which in turn depend upon the realized farm gate prices for the crop.

These factors can all be separated out by applying Monte Carlo simulation, relating stochastic inputs based on likely ranges of water and temperature so as to generate likely unit costs. In the Navatec System a cloud-based project cycle and portfolio management system, there is a considerable emphasis on project design. It provides users with access to linear programming, Markov Models and MCS within an integrated system linked into the project design procedures as a basis for generating optimization, time-based series and Monte Carlo simulation result. A screen shot of an MCS output in the Navatec System is shown below with explanatory annotations added:


By applying linear programming to the feasible ranges to ascertain the gross margins, linear programming can be used to allocate land on a small 10 ha. farm according to the crops to be planted and their contribution to maximization of gross margins. An example from Navatec System is shown below:



Conclusion

Decision analysis has a major contribution to make to agricultural innovation projects. It can ensure a good design in terms of project resilience and sustanability so as to reduce risks of failure during implementation. It can also contribute significantly to the improvement in decision-making during implementation when changes occur. In this way it can enhance the return on investment of lender and donor funds used in the economic development arena as well as help reduce the currently unacceptably high rates of project failures. Unfortunately, the procedures and methods used in decision analysis seldom feature as part of the training of agronomists, economists and those working in the scientific disciplines in the agricultural domain. A lot needs to be done on the side of training. Navatec System handles this by basing its design method on the decision analysis approach. Inputting data on gaps and needs and constraints, feasible solutions and resource requirements in a logic sequential fashion, ends up with appropriate project models that can then be run as analytical simulations.

Training still remains an important requirement but by learning to use the Navatec System it is possible to design optimized projects and secure an ability to make effective decisions in response to change during project implementation.

Click here to access the Navatec System site


  1.  Hector McNeill is an agricultural economist and systems engineer. He is the chief designer of the Navatec System and Director of the George Boole Foundation Limited. He has designed and implemented online systems for over 20 years and develops cloud-based applications based on powerful server side program operations applying ISO and ECMA standard languages. He is the lead international developer of virtual client technology, accumulogs and the Plasma Database. He has a long field experience in agricultural and rural development project assessments, design, management and evaluation on behalf of private companies including Unilever, Mars, Cobec, Express, Prospec, Interbras and Intercomex as well as international agencies including FAO, CBD, EuropeAid, ICO and World Bank, ODA/DfID and the Know How Fund. This work was undertaken in South America, Sub-Saharan Africa, Central and South Eastern Europe and within the EU. He completed well over 300 assessments of project proposals from over 35 countries in development or transition on behalf of Europe Aid in agriculture, rural development, poverty reduction and marginalised communities. He was a member of the Information Technology Committee for European Projects at Manpower Services Commission of the UK Government. He also identified and planned EU wide funding initiatives for 5th Generation and AI-based learning systems applications in agriculture, biomedicine and the environment for the Information Technology and Telecommunications Task Force (ITTTF) at the European Commission. His graduations were from Cambridge and Stanford Universities in agriculture, agricultural economics and systems engineering.

  2.  Examples of models applied to agriculture and natural resources:  McNeill, H. W., "3D production function", Food Research Institute, TP, University of Stanford, 1968; and McNeill, H. W., and Jino, M., "Simulation of 3DPF", CNAE, National Research Council, Brazil, 1969.  McNeill, H. W., "A quantitative model of the interaction of bioclimate and plant biomass production", National Research Council, Brazil, 1970.  McNeill, H. W., "Coffee crop recognition algorithm simulation from high resolution remote sensing data", Plant Production Division, FAO, Rome, 1971.  McNeill, H. W., & Serra, R. "The McNeill-Serra Model" for river basin catchment areas, Mogi Guacu River basin, Institute of Geology, SP, Brazil, 1975.  McNeill, H. W., "The CRESTFILM business process simulation model", Manpower Services Commission, Department of Employment, UK Government, Soft Horizons, Portsmouth 1990.  McNeill, H. W., "Farm production plan optimization system", Online system, Hungarian Agricultural Development Foundation, AFA_Agronet system, Budapest, Hungary, 2000.  McNeill, H.W., "The PAC Model", Online system, SEEL-Systems Engineering Economics Lab, Portsmouth, UK.  McNeill, H.W., "Price Performance Ratio Impacts of Real Incomes", Online system, SEEL-Systems Engineering Economics Lab, Portsmouth, UK.   McNeill, H.W., "Simulations agricultural planning options based on benchmarked data", Online demo at Decision Analysis Intiative 2005-2010, GBF, London, 2008.  McNeill, H. W., "Land Consolidation Co-operative Model - proof of concept", Online system, SEEL-Systems Engineering Economics Lab, Portsmouth, UK, 2014.  McNeill, H. W., "Commodity balance calculator series", Online system, George Boole Foundation, London, UK, 2016.  McNeill, H. W., "Simulation models for optimising project and process designs", Online system, SEEL-Systems Engineering Economics Lab, Portsmouth, UK, 2017.  McNeill, H. W., "Simulation models of optimising agricultural policy designs"Online system, SEEL-Systems Engineering Economics Lab, Portsmouth, UK, 2017.  McNeill, H. W., "Simulation models of optimising proposals for agricultural policy initiative participation.",Online system, SEEL-Systems Engineering Economics Lab, Portsmouth, UK 2017. Navatec System, Cloud project cycle and portfolio management system due diligence design procedures, 2018.


APEurope.org