Difference between revisions of "Decision"

Revision as of 05:46, 27 April 2012

Decision is a special kind of variable: it answer a question like this: "What are different decisions that decision maker X can make in situation Y, and what are different options of each decision?" Decision maker X may be a single individual, a decision making body with several individuals, or a set of separate decision makers, who each can decide about some but not all decisions. In such a situation, an especially interesting part of the related assessment is to look at the interactions of the decisions by different decision makers, none of which can fully control the decision situation.

The decisions have a special role in causal diagrams, as they list such specific structural parts of variables in a causal diagrams that can be modified by a decision maker. Without a decision, the causal diagram describes the business-as-usual (BAU) situation. A decision variable describes how the situation changes if a decision maker chooses an option over another options of a decision.

<section end=glossary />

Question

How should decisions be described in open assessment and what should their structure be?

Answer

Rcode

############### decisions.apply takes a decision table and applies that to an assessment.
## scen is a data.frame that must have columns Decision, Option, Variable, Cell, Change, Result. It can have several variables.
decisions.apply <- function(scen) {
	out <- as.list(unique(scen$Variable))
	names(out) <- as.character(unique(scen$Variable))
	for(variables in unique(scen$Variable)) {
		out[[variables]] <- scenarios.apply(scen[scen$Variable == variables, ])
	}
	return(out)
}


############## scenarios.apply takes the decision table and applies all decisions described on that.
## scen is a data.frame that must have columns Decision, Option, Variable, Cell, Change, Result. It must have only one variable.
scenarios.apply <- function(scen) {
	decs <- data.frame(temp = 1)
	for(decisions in unique(scen$Decision)) {
		temp <- as.character(scen[scen$Decision == decisions, "Option"])
print(temp)
		temp <- data.frame(Options = c(temp, "BAU"))
print(temp)
		colnames(temp) <- decisions
		decs <- merge(decs, temp)
	}

	out <- merge(decs[colnames(decs) != "temp"], get(as.character(scen$Variable[1]))) # This should be updated so that if necessary, the variable is downloaded from Opasnet Base.
	out <- options.apply(scen, out)
	return(out)
}

##### decisions apply the decisions to the cells of a variable. Parameters:
## dec: a data.frame with decisions. It must contain columns Cell, Change, and Result.
## var: a variable data frame.
options.apply <- function(dec, var) {
	for(s in 1:nrow(dec)) { # Each row in decision handled separately
		cell <- strsplit(as.character(dec$Cell[s]), split = ";")
		cell <- strsplit(cell[[1]], split = ":")
		cond <- as.character(dec$Decision[s])
		cond <- var[, cond] == cond # Only the rows with the relevant option.
		for(r in 1:length(cell)) { # All Cell conditions extracted and combined with AND.
			cond <- cond * (var[, cell[[r]][1]] == cell[[r]][2])
		}
		cond <- as.logical(cond)
		if(dec$Change[s] == "Replace" ) {var[cond, "Result"] <- dec$Result[s]}
		if(dec$Change[s] == "Add"     ) {var[cond, "Result"] <- dec$Result[s] + var[cond, "Result"]}
		if(dec$Change[s] == "Multiply") {var[cond, "Result"] <- dec$Result[s] * var[cond, "Result"]}
	}
	return(var)
}

Test code

var1 <- data.frame(c = c("A", "A", "C"), d = c("D", "E", "E"), Result = 4:6)
var2 <- var1
dec <- data.frame(Decision = rep(c("1", "2"), each = 2), Option = c("OptA", "OptB"), Variable = c("var1", "var1", "var1", "var2"), Cell = c("c:A;d:E", "d:D"), Change = c("Replace", "Multiply"), Result = 7:10)
var1
dec
decisions.apply(dec)

Rationale

Previously, it was thought that decision variables are an essential starting point of a causal diagram, and that they should be general and applicable to all possible situations where that causal diagram is used. Now the thinking is different. Decisions are seen as descriptions of plausible options decision makers have in a particular situation. Decisions are implemented in an assessment as scenarios (deliberate deviations from the truth, asking: "What would happen if the truth about the decision maker's action was this?"). It was previously also thought that decisions are probability distributions about what a decision maker will decide, and assessment-specific scenarios are used separately to select the interesting options from the decision distribution in a particular assessment. Now the thinking is different: instead of attempting to describe all the decisions explicitly, the core of a causal diagram does not describe any decisions. It just describes the phenomena that are important for understanding a situation, such as activities, emissions, exposures, and health effects. Decisions that may change the future situation are implicitly included in the probability distributions as uncertainties. Then, decisions are added to that for describing how e.g. activities would change if some decision options would be chosen. Mathematically, this means that one or more variables in a model are conditionalised to reflect the impact of decisions. In other words, decision variables are used as scenarios. What are scenarios, anyway?

A scenario is a deliberate deviation from the truth in a description. In the case of decisions, it is a particular value of some variable that is changed if the decision option at hand is implemented. It can also be a set of particular values of several variables changed by the decision. For policy assessments, often several scenarios are defined and then compared to each other, e.g. if the impacts of a certain policy (measure) is assessed. A particular set of scenarios can be saved and used in several risk assessments. A scenario therefore can be a part of an assessment but is not an assessment itself. In a sense, BAU is not a scenario, it is our best estimate about what will actually happen, implicitly including all plausible decisions by all decision makers.

A decision describes changes in values of other variables. These changes must be within the range of values the variable has.R↻

See also

References

Related files

<mfanonymousfilelist></mfanonymousfilelist>