Impact evaluation Flashcards

1
Q

5 levels of program evaluation

A
  1. Needs Assesment (needs)
  2. Program Theory Assessment (needs, input, output, outcome, impact, long term)
  3. Process/optional evaluation (input, output)
  4. Impact evaluation (outcome, impact, long term)
  5. Cost-benefit analysis
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Theory of change

A

A statement of how inputs being provided lead to intended outcomes and impacts.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Counterfactual

A

A controlgroup representing an identical world without the program

Without controlgroup: Under-estimation of the effect of the program

Cannot attribute causal impact of the program

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

3 Techniques for Impact Evaluation

A
  1. Randomisation
  2. Matching methods
  3. Difference in Difference
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Randomisation, what is random

A

Random sample, treatment+control group, random sample from both groups

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

RCT issues

A

External validity
Hawthorne effects (change behaviour when knowing being observed)
John Henry Effect (control treatments work harder to keep up with treated)
Contaminotion/spillover
Dropout or attrition
Partial Eq (Measure short term effects)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Matching method

A

When randomization is not possible

Propensity Score Matching: find a control group comparable to participant in a large number of observable characteristics: 1. use survey, select large number of characteristics helping predict program participation. Using treated
and untreated estimate the probability of participation p(X ) as function of characteristics → PSM. 2. Match treated to untreated using p(x ). The closer the p(X ) the better
3. The impact of the program is the average difference in outcomes between treated and untreated.

Assumption: unobservable are similar across treated and untreated areas.

Cons: requires lot of data, must be randomisatoin to have similar treaten/untreated

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Differens-in-Difference

A

An alternative is to observe the change over time of non-beneficiaries (when we don’t have random sample).

Subtracts changes.

Parallel trends assumption: without program, the trends between treated and untreated would have been identical.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly