Ex Post Facto Investigations Characteristics and Classification

Introduction

OBJECTIVES

 Know the fundamental characteristic of ex post facto research.
 Identify the validity problems posed by ex post facto research and know the control techniques that can be applied.
 Distinguish between retrospective and prospective research strategies.
 Know the different research designs and analyze their possible sources of confusion.

 Have a perspective of the possibilities of application of ex post facto investigations.

EX POST FACTO = After the events occurred.

LV cannot be intentionally manipulated, nor can subjects be randomly assigned to different LV levels (subjects are selected based on whether or not they possess certain characteristics). Subjects are selected after VI has occurred. They are variables associated with them and, therefore, are already given (gender, age, past experiences, social class, educational program, illness ...). Whoever investigates has no control over the variables, nor can they influence them, because they have already happened, as well as their effects.

They pose VVEE control problems, due to the nature of the variables. We consider two research STRATEGIES:

• -  Retrospective: The causal process has already occurred, it is about looking for the possible causes that have caused them (reconstructing the facts).
• -  Prospective: VI is known but consequences have not been evaluated. Studies on relationships of "pre-existing" variables in people or their determinants

Characteristics of the ex post facto investigation

To study the beneficial effect of nicotine in the performance of cognitive tasks

The fundamental difference between the experimental paradigm and ex post facto research is in control. Subjects are selected with certain IV values ​​(smoking) and equality between the groups with respect to other possible variables cannot be guaranteed..

As there is no intentional manipulation of the VI, but selection of values, causal relationships cannot be established. Only the covariation of the variables with the phenomenon under study can be studied.

REMINDER. Requirements to be able to establish causal relationships:

1) Existence of covariation between LV and RV.

2) LV must precede RV.

3) It should be possible to rule out alternative explanations.

In ex port facto designs, the first requirement is met and, in some cases, the second (prospective designs), but the third is not met. Although we speak of LV (predictor) and RV (criterion), the difference between the two is merely theoretical. In ex post facto research, causal but relational hypotheses are not established.

When the ex post facto design is for group comparison, hypotheses are formulated in which a differential relationship is established between the groups. Subjects with different levels of intelligence (medium and high) on which we study the degree of job satisfaction

Lacking control over VVII, problems arise that affect the internal validity of the investigation. The ex post facto research has less internal validity than the experiment, but it gains in external validity (because the situations are more natural and representative). It is very useful in the applied field. It allows addressing questions that could not be investigated from the experimental method.

Control techniques

In order to clearly establish the relationship between the variables, we should control the possible VVEE that may interfere with the results..

PROCEDURES (Alvarado, 2000):

1. Matching the subjects in other relevant variables that are suspected of influencing RV. Groups of subjects with equal levels of these variables would be formed. It consists of selecting, for each group, subjects with equal values ​​in the relevant VVEE.
2. ANCOVA (Analysis of covariance). Statistical procedure that requires representative samples. It is performed "a posteriori" of data collection. It allows eliminating some sources of unwanted variation (separating the influence of a disturbing variable on RV, thus debugging the effect that IV produces on RV).
3. Introduction of variables related to DV. Multiple VVDDs are used instead of just one. Study on depression (VI) and absenteeism from work (VD1), job satisfaction (VD2)

Classification of ex post facto designs

In ex post facto designs, the researcher selects a sample of people who already have a certain value of the study variables. The possible VVIIs are given and the VVDD can be observed before, after or simultaneously with the observation of the VVII.

Depending on when the DV is measured, a retrospective or prospective strategy is followed.

Retrospective designs

VD and VI have already occurred. Subjects are selected by their RV values ​​and possible causes (VVII) that have caused the response are searched for. If a systematic covariation is found between RV and LV, it can be considered that there is a relationship between both. There are three types:

• -  Simple retrospective design.
• -  Retrospective group comparison design or Retrospective case-control study design.
• -  Retrospective single group design.

Simple retrospective design

1. A group of subjects is selected who possess a certain value of DV.
2. Possible explanations of the phenomenon are sought (VVII).

The DV only takes one value. According to the theoretical postulates existing within the field of study, the possible VVII that could explain the phenomenon are selected (they must precede it). Breast cancer

REQUIREMENTS to be met in the search for possible VVII:

• -  That are relevant to explain the phenomenon under study.
• -  That precede the DV Utility ...

• -  Exploratory if no information is available on the variables that could explain the phenomenon.
• -  Functional if the problem to be investigated is too complex to be studied with another type of design. DATA ANALYSIS: If the DV only takes values ​​of occurrence, we can analyze the percentage of cases that each IV shares with the DV. INTERNAL VALIDITY is very weak because it is difficult to rule out alternative explanations (differentiate the relevant variables from those that are not).

Retrospective group comparison design = Retrospective case-control study design

It consists of comparing a group of subjects selected for possessing a certain characteristic (cases) with another group of subjects who do not possess it (controls). These groups are compared with respect to a series of possible VVII that are considered relevant for the appearance of this characteristic..

The only DIFFERENCE between the groups: That one has the characteristic that is interesting to investigate and the other does not (DV takes two values). They must be equivalent in all other factors. This can be controlled by pairing.

ANALYSIS OF DATA:

-  Compare the proportions of the groups.

-  Study the correlations between DV and IVI.

-  Study the relationships between the VVII, the less relationship there is between them, the easier it will be to establish an overall vision.

Greater INTERNAL VALIDITY compared to the simple retrospective design, although there is still the possibility of

numerous VVEEs that could lead to a misinterpretation of the results.

Retrospective single group design or retrospective correlational design

OBJECTIVE: Study of the relationships between the variables. No comparisons are made between the different LV levels with respect to RV, correlations are established between the variables (while in the previous, simple retrospective or group comparison designs, the objective is similar to that of the experimental logic: to find possible VVII that explain DV).

If what you want to study is the covariance or correlation between DV and VI, it is convenient that all the possible values ​​of the two variables are well represented. This is achieved by increasing the size and representativeness of the sample. INTEREST: Getting all the possible values ​​of the variables involved to be well represented in a single group with which to work.

PROCESS:

1. Select the group of subjects that guarantees the representation of the range of DV values.

2. Determine which could be the relevant VVII and measure them in the entire sample.

DIFFERENCES with respect to the simple retrospective design:

• -  DV takes different values.
• -  We work with very large samples (maximum representativeness of the population under study to ensure that the range of LV and RV values ​​is well represented).

DATA ANALYSIS: Correlations between variables.
Greater EXTERNAL VALIDITY for working with very large samples (greater possibilities of generalization). Regarding INTERNAL VALIDITY, there can be no guarantees that the occurrence of the DV will be after the presence of the IVI.

Prospective designs

Only VI is given. Its possible consequences have not yet been evaluated. Subjects are selected for having certain LV values ​​whose possible influence we want to investigate. Self-esteem and school performance

They have GREATER INTERNAL VALIDITY than retrospective designs. There are three types:

-  Simple prospective design.

-  Complex prospective design.

-  Single-group prospective design.

We are approaching a relationship in which VI precedes RV, since this has not yet occurred.

1. Select us who participate for their value in the VI
2. We measure the DV

Simple prospective design

We have a single VI and we select the subjects by their values ​​in it (value selection variable) forming, at least, two groups, one for each value. Subsequently, the RV is measured.

LV (soccer practice -usually, not usual-) + RV (heart rate)

Design similar to the experimental design of two random groups, with the following DIFFERENCES:

-  The VI is already given.

-  Control over possible VVEEs is much less due to selection.

-  Limitations of the interpretation of the results (due to control problems).

-  Difficulty safely ruling out alternative explanatory hypotheses.

CONTROL techniques:

- Pairing.

-  Estadistic control.

-  Introduction of variables related to DV.

ANALYSIS OF DATA:

• -  If the VI takes only two values: Difference of measurements for two independent samples.
• -  If more than two groups are selected: One-way ANOVA with independent groups or samples.

Complex prospective design

When an EV is suspected of influencing the results, its effect can be controlled by incorporating it into the investigation as VI. Study the effect of more than one VI on RV.
VI1 (soccer practice -usually, not usual-) + RV (cardiorespiratory rate) + VE (VI2, overweight BMI -<25, >25-) It has the structure of a 2 × 2 factorial experimental design but, in this case, the VIs are of value selection

You have to form as many groups as possible combinations of the levels of VI can be made. 4 groups DATA ANALYSIS: Two-factor ANOVA with independent samples.
It has GREATER INTERNAL VALIDITY than the simple prospective design since it controls an EV that is

you suspect that it may influence the results. Although it continues to have validity problems because the subjects are selected for their values ​​in the VVII and this can give rise to other VVEE associated with them. The more variables we include, the greater internal validity, but also the more complexity in the design (the formation of groups with all possible combinations may become unfeasible).

Single-group prospective design

Same logic as retrospective single group design, but ...

• 1. Select a sample in which all the VVII values ​​are present.
• 2. All VVII that are considered to be related to the DV that we want to investigate are measured.
• 3. DV is measured.
• 4. The relationship between the variables is studied.

Study of the influence of personal and psychosocial variables on the academic performance of student athletes belonging to private schools for students of color.

It has a GREATER EXTERNAL VALIDITY because the number of VVIIs to be considered in the investigation is increased, but a much larger sample is required.

It also has GREATER INTERNAL VALIDES since VVII are measured first and then VD. Thus we know that LV precedes RV (although there may still be multiple alternative explanations for the effect of VVEE that have not been considered in the design).

PURPOSE:

• Descriptive: To study variables that covariate with the phenomenon of interest and are useful when the research problem is very complex or little is known about it..
• Predictive: If there is greater knowledge about the research problem. It works with all the scores. It is intended to find a function that relates the scores in the predictor variable (VI) with the criterion variable (VD).
• Using simple linear regression analysis, from a LV (X) score, the value of DV (Y) can be predicted.
• By means of multiple linear regression analysis, it is intended to forecast by more than one VI (X1, X2… Xn), the value of the DV (Y) and find out the variables that are better as predictors.

Evolutionary designs

They use age as VI. OBJECT: Study of RV changes as a consequence of age. Widely used in Developmental Psychology, which studies differences in behavior during the development process. To what extent do attitudes towards sex change with age??

a) Longitudinal evolutionary design: Records data from the same sample of subjects on successive occasions over time. Study the development of people over a long period.

• -  THREATS: By always evaluating the same sample, the cohort threat is controlled, but not the history (the effect of age could be confused with the effect due to the time the test is performed), nor the experimental mortality (at require long follow-up, many people may drop out).
• -  ANALYSIS OF DATA:

o Difference of measurements for related samples, when in the design we only have two moments in time of the DV measurement.
o One-way ANOVA with repeated measures, when we have DV measurements collected at more than two time points.

b) Cross-sectional evolutionary design: Data collection is carried out in a single time point to samples of people of different ages. It studies the differences by age, although in the field of Evolutionary Psychology they are used to study changes produced with age (under the assumption that the differences found between the different age groups are due to the different moments of the development process that they share participants). It is less expensive because it is more limited in time.

• -  THREATS: Due to its temporal limitation, the threat of experimental mortality is controlled but, sometimes, the groups are not comparable because, in addition to age, there may be multiple variables that could explain the differences found in DV (matching or introduction of variables related to DV to improve internal validity). The cohort effect (are the differences due to age or generation?) May reduce internal validity. Study of development in girls and boys of 3, 4, 6 and 8 years
• -  ANALYSIS OF DATA:

o Difference of measurements for independent samples, or its non-parametric equivalent, if we are studying only two age groups.
o One-way ANOVA with independent groups, or its non-parametric equivalent, if we have more than two age groups. c) Sequential evolutionary design:

• -  Longitudinal sequential evolutionary design: When, in addition to age, you want to study the cohort effect. Longitudinal comparisons would be made within each cohort (two or more cohorts can be studied at two or more time points).
• -  Cross-sectional sequential evolutionary design: Comparisons would be made at different moments of the investigation. More expensive than longitudinal (requires more time and more participants, sometimes unfeasible) and with a higher probability of non-random loss of participants (which could cause greater bias); This is why they are not often used in Developmental Psychology ADVANTAGES: Combining longitudinal and cross-sectional evaluations can control for cohort effects and historical effects.

DATA ANALYSIS: Mixed design or split-plot = ANOVA of two factors with repeated measures in one of them (in which the different time points are a factor of repeated measures and the cohort effect is a factor of independent measures).

Evolutionary designs, when using age as VI, can be affected by THREATS to INTERNAL VALIDITY related to selection (cohort, historical, cultural factors…). Alternatively, a "true VI" (intentionally manipulated) can be entered. We would thus have a factorial design with a VI for selection of values ​​and a VI for intentional manipulation and we could study the effects of each variable separately and jointly..

Use of ex post facto designs in applied research

They are very useful to approach the study of certain problems that could not be investigated from other methods due to their complexity and the nature of the variables involved. They serve to explore and generate hypotheses that suggest new research.

Of great application in CLINICAL PSYCHOLOGY for:

• -  Establish diagnostic categories or make predictions about diagnosis and therapy.
• -  Do epidemiological investigations.

The TYPES of studies most frequently used are:

1. Descriptive studies: When little is known about the occurrence, natural history or determinants of a disease

 Methodology: Survey.

 Objectives:

-  Estimate the frequency or trend of a disease in a particular population.

-  Generate specific etiological hypotheses.

2. Etiological studies: When enough is known about the disease and there are specific hypotheses.

 Objectives:

-  Identify risk factors for the disease, estimate their effects.

-  Suggest possible intervention strategies. Other fields of application are:

-  The EDUCATIONAL FIELD to study variables related to school performance or success. Self-concept, gender, aptitude, cultural differences, family environment ...

-  DEVELOPMENTAL PSYCHOLOGY (evolutionary designs) to establish relationships between age and the use of certain strategies. Learning, memory, linguistic functions ...