Panel data

In panel data is two-dimensional data that is collected as part of a panel study. Panel data are to be distinguished from cross-sectional data , in which the units are recorded at a single point in time, and time series data , in which a single unit is observed over several time periods. In addition to the same units of investigation (e.g. individuals, households or companies) in the panel and the multiple implementation of the survey (e.g. monthly, annually or biennially), the third characteristic mentioned is that the same or at least the same content asks for or be collected.

The panel data is the statistical analysis of panel data under the panel research. Although a panel is still observational (there is no intervention), a key goal is to get as close as possible to causal analysis.

Panel data is organized in tables made up of rows and columns. There are two main formats. In the wide format, one line corresponds to an examination object and the measured values of the times are appended as columns. This is different with the long format, in which a line is added for each observation carried out, i.e. the same object appears in several lines (cf. wide format and long format ). If each person-year combination is viewed as an individual observation, one speaks of pooled data .

evaluation

advantages

Compared to cross-sectional surveys , in which all relevant variables are recorded simultaneously with a single measurement, panels have the advantage that causal questions can be dealt with much better because changes at the individual data level can be determined on the same object. However, the causal conclusion is not particularly valid due to many non-excluding influencing variables, as is the case with all quasi-experimental studies.
Compared to trend surveys , in which new participants are selected for each study, panel studies have the advantage that the behavior of the respondents can be observed over several measurement times. This enables time series studies to be carried out on the level of individual data, from which causal relationships between several variables recorded in the panel can be inferred. Of course, the panelists can also take part in additional ad hoc investigations.
Compared to classic ad hoc methods, the household / consumer panels can use classic screening methods to identify (special) target groups that are often difficult or impossible to find (e.g. households that have bought a certain product in the last quarter ) - select representative and thus question market research economics. The basis for this is the (purchasing) data from the continuous tracking that takes place in the panels. Since the data is available for each household, target groups relevant to the survey can be selected representative “at the push of a button”.

disadvantage

Constancy of the measuring instruments : The socio-cultural change leads to a change in the meaning of terms. It can therefore happen that the same question 10 years later no longer measures the same characteristics of the same construct. At the same time, new, in some cases better, social science measurement and selection processes are being developed. This results in a conflict between the endeavor to use measuring methods with the highest possible quality and to ensure the constancy of the measuring instruments.
Panel mortality : In the course of panel examinations it often happens that participants e.g. B. due to illness, relocation or loss of motivation to participate for a long time not actively participating in the examination or dropping out. Those participants who were active in the panel during the entire study period are therefore of particular interest for the study. This subset of the panel is called the continuous mass . The continuous mass becomes smaller the longer the investigation period is. This problem is known as panel mortality or panel mortality. It is particularly problematic because it can lead to systematic distortions: Often the eliminated persons systematically deviate from the subjects remaining in the panel with regard to one or more characteristics. In order to counteract the problem of panel mortality, attempts are made to keep the failure rate low through panel maintenance and to select a sufficiently large sample.
Panel effects : The repeated questioning can change the objects examined. It can happen that the interviewed persons develop their attitudes towards a certain topic only after the interview. In addition, existing attitudes and behaviors can be changed or consolidated through repeated questioning. Both distortions are combined as panel effects. In addition, knowledge of the examination process can change the response behavior compared to the initial survey. In interview studies, the development of friendly relationships between interviewers and respondents can also influence response behavior.
Panel rigidification : Even if the survey unit remains the same, personal changes in the respondents, such as B. the birth of children or the professional or financial advancement lead to the fact that the panel loses its representativeness for the population.
Panel selection effects describe the systematic distortions of the results of a panel study due to selective recruitment processes and selective failures. This problem occurs especially when the sample is formed from people who voluntarily declare their willingness to participate in the panel investigation, as happens with access panels in commercial market research.
Personnel and financial effort : Another disadvantage of panel research, albeit not of a methodological nature, is the higher personnel and financial effort involved in managing and supervising participants.

Classifications and properties of panels

Classification according to the examination object

In operation panels , the same companies are asked repeatedly. The most important establishment panel is the IAB establishment panel .
In the case of person panels , the same people are questioned repeatedly. Depending on the objective of the study, the panel can be an individual panel that deals with individuals, a household panel that interests everyone in the household, or a specialized panel such as B. Motorists panel, children panel etc. The most important person panels in the Federal Republic are
- the “ Socio-Economic Panel ” (SOEP) and
- the microcensus (this is also collected in Austria and Switzerland). 1% of private households in Germany are surveyed every year , which corresponds to around 390,000 households with 830,000 people.
The cohort study can be viewed as a special panel of people. Its specialty is that all persons belong to a sample of the same cohort , i. This means that a specific biographical event (e.g. birth, school enrollment, marriage or divorce) occurred in your résumé at approximately the same point in time. The most important cohort study in the Federal Republic of Germany is the “ National Education Panel” (NEPS).

Division according to panel design

In the simple panel study, the entire sample is questioned in each survey. In order to reduce the influence of panel mortality and panel effects, further panel designs were developed. After the implementation, one can, for example, distinguish between online panels . Large studies are also conducted using mixed methods, such as Computer Assisted Telephone Interview or Computer Assisted Telephone Interview.

Balanced data (“balanced panel data”) represent an ideal-type data set in which all data for all points in time are available for all individuals. They are a little easier to describe statistically. In reality, the data is mostly incomplete, which is why we speak of unbalanced panels. The use of unbalanced data is not a problem for the models under consideration, provided that the data are randomly missing and there are enough successive observations. Systematically missing observations, on the other hand, lead to a selection bias and make special estimation procedures necessary. Another advantage is the enlargement of the database. Usually unbalanced (“unbalanced”) panel data are available, in which individuals leave the survey earlier or join later.
In the alternating panel, the sample is divided into subgroups that are interviewed alternately. For example, only the first subgroup could be surveyed in the first survey, only the second in the second survey, the first again in the third survey, and so on. In this way, the stress on the test subjects is reduced, and the longer time intervals reduce the risk of panel effects. However, this procedure should only be used with sufficiently large samples.
Another design variant is the rotating panel. Here one subgroup is eliminated from each survey wave and is replaced by a new one. However, this means that each subgroup takes part in the survey differently. The group that only participated once must be excluded. The panel mortality of the group that took part in all surveys is particularly high.
The divided panel ( "split panel survey") has the advantage that for each survey wave a control group is present. In this way it can be determined whether panel effects are occurring and how strong they are. To do this, you form two groups. A simple panel study is carried out with the first group (see above). The second group goes through either a rotating panel or a repeated cross-sectional examination.

The following shows sample data for a balanced and unbalanced panel. On the right hand side, data is not available for every person at all times. Both tables show panel data in long format.

balanced panel:		unbalanced panel:
${\ displaystyle {\ begin {matrix} \ mathrm {person} & \ mathrm {year} & \ mathrm {income} & \ mathrm {age} & \ mathrm {gender} \\ 1 & 2001 & 1300 & 27 & 1 \\ 1 & 2002 & 1600 & 28 & 1 \\ 1 & 2003 & 2000 & 29 & 1 \\ 2 & 2001 & 2000 & 38 & 2 \\ 2 & 2002 & 2300 & 39 & 2 \\ 2 & 2003 & 2400 & 40 & 2 \ end {matrix}}}$		${\ displaystyle {\ begin {matrix} \ mathrm {person} & \ mathrm {year} & \ mathrm {income} & \ mathrm {age} & \ mathrm {gender} \\ 1 & 2001 & 1600 & 23 & 1 \\ 1 & 2002 & 1500 & 24 & 1 \\ 2 & 2001 & 1900 & 41 & 2 \\ 2 & 2002 & 2000 & 42 & 2 \\ 2 & 2003 & 2100 & 43 & 2 \\ 3 & 2002 & 3300 & 34 & 1 \ end {matrix}}}$

More panels

literature

Martin Günther, Ulrich Vossebein, Raimund Wildner: Market research with panels. Types, survey, analysis, application. 2nd completely revised edition. Gabler, Wiesbaden 2006, ISBN 3-409-22244-8 , pp. 1-72.
Rainer Schnell , Paul B. Hill , Elke Esser: Methods of empirical social research. 7th completely revised and expanded edition. Oldenbourg, Munich et al. 2005, ISBN 3-486-57684-4 , pp. 237-245.
Andreas Diekmann : Empirical social research. Basics, methods, applications (= rororo. Rowohlt's Encyclopedia 5567818). 1st edition, completely revised and expanded new edition (1st edition of the new edition). Rowohlt-Taschenbuch-Verlag, Reinbek near Hamburg 2007, ISBN 978-3-499-55678-4 , pp. 304-312.

Individual evidence

↑ Baur, N., & Blasius, J. (Eds.). (2014). Manual methods of empirical social research. Wiesbaden, Germany: Springer VS. P. 925
↑ Microcensus. To the survey. In: destatis.de - Methods and Procedures. Federal Statistical Office Germany, August 31, 2007, accessed on October 19, 2013 .

[1] Baur, N., & Blasius, J. (Eds.). (2014). Manual methods of empirical social research. Wiesbaden, Germany: Springer VS. P. 925

[2] Microcensus. To the survey. In: destatis.de - Methods and Procedures. Federal Statistical Office Germany, August 31, 2007, accessed on October 19, 2013 .