Disruptive factor

The articles spurious correlation , intervening variable and disruptive factor overlap thematically. Help me to better differentiate or merge the articles (→ instructions ) . To do this, take part in the relevant redundancy discussion . Please remove this module only after the redundancy has been completely processed and do not forget to include the relevant entry on the redundancy discussion page{{ Done | 1 = ~~~~}}to mark. Zulu55 ( discussion ) _{_ignorance} 20:27, 10 Dec. 2013 (CET)

Disturbance factor (not to be confused with disturbance parameter , or disturbance variable , also called third variable ) is a term from the empirical field in experiments. It is all those factors that can influence both the dependent variable and the independent variable and are not manipulated. These can be characteristics of test subjects or external factors. Third variables are to be seen as alternative or competing explanations for the initial hypothesis of the research problem. There are special techniques for controlling disruptive factors.

Under a confounder ( Engl. For: 'nuisance' of Latin confundere : confuse, mix together water) or Germanized Konfundierungseffekt (see also ". Confused is understood within") epidemiological causes a problem, the two factors of observational studies, namely exposure as well as the end point. A confounder is a variable that influences the occurrence of a risk factor and the observed endpoint at the same time.

The observed exposure is not the sole cause of the observed effect - this is at least partially caused by a confounder.

species

In the social sciences in particular, it is necessary to examine the disruptive factors more closely. The variance is used for this. The variance is a measure of dispersion for formal representation. The variance is the square of the standard deviation. Mathematically, the variance is the mean of the squared deviations of all individual values from the overall mean.

Conditions for confusion

An experiment tests whether a random variable has an influence on a random variable . If, however, in addition to the known variable , an unknown disturbance variable also influences the random variable , then one speaks of confounding. If a third variable influences two random variables, then the causal interpretation of the effects is falsified. Confusion is one possibility of this corruption. It depends on two conditions: ${\ displaystyle X}$ ${\ displaystyle Y}$ ${\ displaystyle X}$ ${\ displaystyle W}$ ${\ displaystyle Y}$

A disturbance variable is stochastically related to the independent variable . ${\ displaystyle W}$ ${\ displaystyle X}$
Disturbing variable changes the regressive relationship between the dependent variable and the independent variable . ${\ displaystyle W}$ ${\ displaystyle Y}$ ${\ displaystyle X}$

Expressed mathematically, a regression is confounded if: ${\ displaystyle \ operatorname {E} (Y \ mid X)}$

The events and are stochastically dependent and ${\ displaystyle X \! \, = x}$ ${\ displaystyle \! \, W = w}$
${\ displaystyle \ operatorname {E} (Y \ mid X = x) \ neq \ operatorname {E} (Y \ mid X = x, W = w)}$

Sources of disruptive factors

Internal and external validity for experiment and quasi-experiment

Sources that affect internal and external validity :

Occurrences in the meantime (events that additionally influence the dependent variable in addition to the stimulus, e.g. "Black Friday")
Maturation processes ("intrapersonal" processes that are independent of the stimulus , e.g. development of a small child)
Subject motivation, for example in the form of the effect of social desirability
Effects of the special test conditions and methods ( testing effects ), see reactivity (social sciences)
Aid (change in the measuring instrument), this can also be the involuntary change in the experimenter's gestures
Skewed selections and failures (difference between control and experimental groups not only in terms of the stimulus but also in other characteristics that influence the dependent variable)
Experimenter Effects.

If disruptive factors and stimuli are mixed, one speaks of a confusion.

Common risk factor

If one examines the connection between tobacco smoking and liver cirrhosis or hepatocellular carcinoma , a clear association can be established. However, there is no biological connection: Smoking does not lead to liver cirrhosis. Rather, many drinkers are also smokers (statistical association based on an overarching common cause ( addictive personality )) and alcohol consumption is an independent risk factor for liver cirrhosis. In this example, addiction personality and alcohol confounders would be in the context of measuring the effect that smoking has on the outcome of cirrhosis of the liver.

Suchtpersönlichkeit  →   Alkohol
        ↓                   ↓
      Rauchen    →    Leberzirrhose

Testing of confusion

Measurement results are ideally characterized by internal validity . This means that the dependent variable is actually explained by the research approach including the independent variable. If the measurement is influenced and distorted by an interfering variable, i.e. if there is a confounding, the internal validity is no longer given.

It must be expected that the independent variable will be influenced by other variables that also influence the dependent variables. This superposition makes an exact breakdown of the influences of dependent variables difficult or impossible.

In order to find out whether there is a confusion and, if necessary, to weaken it, the model must be checked. There is no specific test for confusion, however, as test problems are usually tested asymptotically. Large samples are necessary for this and statistical inaccuracies with regard to the level of significance are to be expected. However, the empirically often hard to meet requirement is that the disturbing influences can be defined, separated from one another and measured reliably and validly (see the methodologically hardly possible differentiation between different response tendencies in psychological diagnostics ).

Instead, the conditions for confiscation are used. First of all, a potential disturbance variable that can be responsible for the confusion must be found. The next step is to test the two conditions for confusion. On the one hand, there is the stochastic dependency between the independent variable and the disturbance variable . The events and must be stochastically dependent . This can be checked, for example, with a chi-square test . If the first condition is met, the difference with regard to the expected values of the model can then be checked. If the relationship between the independent variable and the dependent variable changes as soon as a potential confounding variable is added to the model , the second condition for confounding is also met . ${\ displaystyle (X)}$ ${\ displaystyle (W)}$ ${\ displaystyle X = x}$ ${\ displaystyle W = w}$ ${\ displaystyle (X)}$ ${\ displaystyle (Y)}$ ${\ displaystyle (W)}$ ${\ displaystyle \ operatorname {E} (Y \ mid X = x) \ neq \ operatorname {E} (Y \ mid X = x, W = w)}$

Confounding is therefore present and can be tested if both conditions are met, i.e. independent variables and confounding variables are stochastically dependent and the expected values of the model with and without confounding variables are each different in size. ${\ displaystyle (X)}$ ${\ displaystyle (W)}$ ${\ displaystyle (W)}$

Control of disruptive factors (third variable control) and avoidance

If a confusion of two variables is only established afterwards and the confounding variables were not recorded in the experiment, the whole experiment becomes unusable, since it is no longer possible to unambiguously deduce the dependent variable from the independent variable.

Randomization is an effective way of preventing confusion in advance . The test subjects are assigned to the various test conditions using a random procedure. This ensures that there is no systematic connection between the dependent variable and possible confounding variables, such as certain personal characteristics. However, randomization is only possible in real experiments in which the assignment of the persons to the respective treatment groups is under the influence of the investigator. For all other survey methods, such as B. quasi or field experiments or pure observation methods, a randomized assignment of the test subjects is not possible and the risk of confusion is therefore in principle present.

The use of a randomized assignment of the test subjects is only effective in the case of large samples , since an equal distribution within the individual groups can only be assumed if the sample size is sufficiently large. In reality, however, the samples are often rather small due to economic considerations or other practical reasons and therefore randomization does not make sense. In these cases, for example, by keeping the possible confounding variable constant, an attempt is made to prevent confounding and thus a distortion of the result. Another possibility is balancing, in which the different manifestations of the possible confounding variable are evenly distributed among the test groups.

In the case of test forms in which it is not possible to influence the sample composition beforehand, it is important that the test leader thinks about possible confounding variables in advance and includes these in the investigation. This is the only way to check afterwards whether two variables have been confounded, and the result can be taken into account using statistical control techniques.

In experiments there are techniques for controlling confounding factors . These techniques are particularly important in social sciences. In the experiment, test and control group (s) can be formed, which serve to eliminate the influence of test person characteristics that can act as disruptive factors. There are two methods of forming groups:

Randomization means that the test subjects are assigned to the experimental and control group at random. This ensures that the differences between the test groups are averaged out with a sufficiently large sample. Randomization rules out the possibility of systematic biases in the results due to the division of the test subjects into experimental and control groups.
Matching or parallelization refers to processes for forming groups that are homogeneous with regard to one or more disruptive factors. For example, if a teaching method is to be evaluated , two groups of pupils that are as similar as possible in terms of their grades can be formed by parallelization.

External factors can also be controlled in laboratory experiments:

Elimination refers to the elimination of possible interfering variables. Your aim is that the test subjects are not affected by any other factors besides the independent variable. In order to ensure that the test subject is not influenced by external events, experiments can be carried out, for example, in windowless, sound-insulated cabins.
Keeping constant : To ensure that the observed effect is due to the variation of the independent variables, an attempt is made to keep all other factors constant. Since the natural brightness fluctuates from day to day and in the course of the day, z. B. Experiments on visual perception can be carried out in a laboratory that is equally lit for all experiments.

Examples

Hawthorne Effect

A famous example of the occurrence of confusion is the so-called Hawthorne experiment from the 1920s. The Hawthorne effect that occurred in these group-based observational studies in the USA describes the influence of confounding variables on an experiment.

In the Hawthorne works (Illinois, USA), an industrial company producing telephones, the work environment was specifically changed in several test runs in order to motivate employees to produce higher quantities. In addition to better lighting, the wall color was changed or the room temperature increased in the subsequent steps. Immediately after every change, an increased production rate could be observed for a short time, which however returned to the starting level after a few days. Thus, not a single change in the working environment led to a permanent increase in the production rate. Rather, there was a mixture of different variables or the appearance of a third variable (disturbance variable), i.e. confusion. The increased work performance could thus be explained by a short-term increased work motivation and not by the improvement of the lighting, the change in the wall color or the increase in the room temperature.

Crozby also calls this phenomenon the “third variable problem”. He states that there does not have to be a direct connection between the variables leisure activity and inner restlessness, but that a higher income may allow more time for extended leisure activities. If income is the determining variable, no cause-effect chain can be established between the examined variables of leisure activity and inner restlessness. The relationship was influenced by a third variable that provides an alternative explanation for the effects observed.

literature

Rainer Schnell among others: Methods of empirical social research. Munich 1994, ISBN 3-486-22728-9 .
Jürgen Bortz, Nicola Döring: Research methods and evaluation for human and social scientists. 4th, revised edition. Springer-Medizin-Verlag, Heidelberg 2009.
Paul C. Cozby: Methods in behavioral research. 10th edition. McGraw-Hill Higher Education, Boston 2009, ISBN 978-0-07-337022-4 .
O. Huber: The psychological experiment. An introduction. Hans Huber Verlag, Bern 2000.
Ingeborg Kittmann In: Giselher Guttmann (Ed.): General Psychology - Experimental Psychological Basics. 2nd Edition. WUV University Press, Vienna 1994.
Christof Nachtigall, Ute Suhl, Rolf Steyer: Introduction to Confounding Analysis. In: Methevalreport. 2 (1), 2000.
KW Schaie: Methodological problems in the descriptive developmental psychological investigation of adult and old age. In: PB Baltes, LH Eckensberger (Ed.): Developmental Psychology of the Lifespan. Klett-Cotta, Stuttgart 1979.
Rainer Schnell, Paul B. Hill, Elke Esser: Methods of empirical social research. 8th, unchanged edition. Oldenbourg-Verlag, Munich 2008, ISBN 978-3-486-23489-3 , p. 208.
Winfried Stier: Empirical Research Methods. 2., verb. Edition. Springer, Berlin / New York 1999, ISBN 3-540-65295-7 .
MR Waldmann: Experiments and causal theories. In: D. Janetzko, HA Meyer, M. Hildebrandt (Hrsg.): The experimental psychological internship in the laboratory and WWW. Hogrefe, Göttingen 2002, pp. 13–42.

Individual evidence

^ KJ Rothman, S. Greenland, TL Lash: Modern Epidemiology. Lippincott Williams & Wilkins, 2008, ISBN 978-0-7817-5564-1 , pp. 130-137. ( limited preview in Google Book search)

[1] KJ Rothman, S. Greenland, TL Lash: Modern Epidemiology. Lippincott Williams & Wilkins, 2008, ISBN 978-0-7817-5564-1 , pp. 130-137. ( limited preview in Google Book search)