General Elimination Methodology

Synonyms: 
GEM

General Elimination Methodology has two stages:

1. Identify possible explanations.  One of these possible explanations is that the program (or project etc) produced the outcomes and impacts that can be observed.  At this stage you should identify as many possible alternative explanations using a combination of options such as:

  • Key informant interviews - well-informed local people might know about historical events, local conditions and/or other programs that could have produced the results
  • Previous evaluations and research - these might have identified other factors that can produce the results
  • Brainstorming

2. Gather and analyze data to see if the possible alternative explanations can be ruled out.

For example, let's imagine you're evaluating a project that aims to support farmers to apply fertilizer to their winter crops to increase production and hence their income and well-being.  If you had data that showed they had had an increase in their annual income, this might be because the program had been effective or this might have been caused by something else.

You would start by possible alternative explanations for the increase in annual income. For example, maybe there was a drought in other areas, so local farmers were able to get a higher price for their crops, even though they had not produced more.  Or maybe their increased income had been from the summer crops.

Then you would gather and analyze data to see if these possible alternative explanations could be ruled out.  For example, if you had some data about local prices which showed they had been stable, you could rule out increased prices as the reason for increased income. If you had information about when people's income had increased, you might be able to rule out income from their summer crops.

This page is a stub (a minimal version of a page). You can help expand it. Click on Contribute Content or Contact Us to suggest additional resources, share your experience using the option, or volunteer to expand the description.

Updated: 5th November 2014 - 4:41am

Comments

rickjdavies's picture
rick davies

When I first heard of this method I thought it would be difficult to use because the number of alternate explanations for an outcome of interest is potentially unbounded, limited only by the amount of time you have and the number of participants available to generate ideas. Drawing a boundary around the set to be explored would seem to be a fairly arbitrary process, so the results of the exercise would always be open to question.

But since then I have been exploring the use of QCA and Predictive Analytic methods I am now wondering if this problem is at least partly resolvable. Both sets of methods make use of data sets of cases (in rows) with attributes (in columns). The number of potential "explanations" for an outcome is equal to 2 to the power of the number of attributes (assuming their presence absence is coded in binary form). It is a bounded set and one where all possibilities can practically be explored by both QCA and predictive analytic approaches.

Of course there is still the question of how you choose (and limit) the set of attributes to include in the data set. There are two approaches to this task. One is theory informed, used by QCA. The other is more data informed, used by predictive analytic people. I find both useful but I will focus on the latter here. A data informed approach can seek to choose those attributes which maximise the diversity of cases (i.e. no cases with identical sets of attributes) but which also maximises the proportion of all possible configurations captured by a set of cases. These two requirements are in tension. A small number of attributes means  we are more likely to have full coverage of all possible configurations, but it also means greater likelihood of duplicate cases. Fortunately the specific choice of attributes for a given set of cases can be optimised (i.e. do well on both criteria) by using add-ins like Evolver in Excel.

Add new comment

Login Login and comment as BetterEvaluation member or simply fill out the fields below.