Experimental Design and Errors Flashcards
Quasi-exp Design: Non-random Control Trial
When a sample cannot be randomly selected (ethical reasons) the subjects self-report to either the treatment or control group. This means that the group allocation of the participants is already determined prior to the start of the experiment. Apart from the lack of randomisation of subjects to groups, everything else would be the same as in an RCT - groups are still compared however you cannot say whether this is due to group difference or another factor.
Quasi-exp Design: Interupted time series
A time series is a continuous sequence of observations on a population, taken repeatedly (normally at equal intervals) over time. In an ITS, a time series of a particular outcome of interest is used to establish an underlying trend, which is interrupted by an intervention at a known point in time. The hypothetical scenario under which the intervention has not taken place and the trend continues unchanged, is referred to as the counterfactual - a scenario that provided a comparison of the evaluations of the impact of the intervention by examing any change occurring in the post-intervention period. It requires a clear differentiation between the pre- and post-intervention period. Works best with short-term outcomes that are expected to change either relatively quickly after the intervention is implemented or after a clear lag.
Quasi-exp Design: One group Pretest Posttest
One of the most frequently used Quasi-exp - a single group of research participants is pretested, given some treatment or IV manipulation, then posttested. If the pretest and posttest scores differ significantly, then the difference may be attributed to the IV, but as the research design is not strictly experimental and there is no control group, this inference is uncertain and the difference may be due to extraneous variables such as order effects.
Quasi-exp vs. true experiment SIMILARITIES
Study subjects are subjected to some type of treatment or condition
Some outcome of interest of measures
The researcher tests whether the difference in the outcome is related to the treatment.
Quasi-exp vs. true experiment DIFFERENCES
In a true exp, participants are randomly assigned to either treatment or control groups, whereas they are not assigned randomly in quasi exp.
Quasi exp control and treatment groups differ in terms of experimental treatment they receive and in other unnown ways, thus, the researcher must try to statistically control for as mant of these differences as possible.
As control is lacking in Quasi, there may be several ‘rival hypothesis” competing with the exp manipulation as explation for obbsereved results.
Quasi-exp vs. true experiment
Both designs purpose is to examine the cause of certain phenomena. True exp, all factors that might affect the phenomena of interest are completely controlled for, the preferred design. Often not practical or possible to control all the key factors, so it becomes necessary to implement a quasi-exp research design.
Alternate explanations: History Effect
ANy external event that occured between the pretest and posttest that causes the chance in measure and not the intervention. The longer the study takes, the more likely it is, that history effect may become a threat to the validity of the study.
Alternate explanations: Testing Effects
Effects of taking a test on the outcome of taking a second test. The fact that people were tested, not the treatment, causing change in behaviour.
Alternate explanations: Maturation Effect
Process within subjects which act as a function of the passage of time. Change simply with time and not due to exposure to the treatment or intervention.
Alternate explanations: Regression Towards The Mean
Measurement always in a zig-zag fashion (not steady). Goes from one extreme to another on alternating measurements taken.
Alternate explanations: Instrumental Decay
Gradual loss of accuracy of measurement. Usually a problem when using frequency data. Effect not due to intervention but the fact that fewer events were recorded.
Random Error
Always present in a measurement. It is caused by Inherently unpredictable fluctuations in the readings of measurement apparatus or experimenters recordings. Measurable values are therefore inconsistent when repeated - causes randomly by chance.
Random errors are self-canceling - some scores go up, while others go down.
Bigger sample size = less chance of random error,
Systematic Error
Not determined by chance but introduced by inaccuracy. If the cause of systematic errors can be identified, they can, therefore, be eliminated.
Systematic error is when there is bias because of how the sample is collected - eliminated with the design.
