Flashcards
ERM 10-12
Flashcards do conjunto (89)
- 1Statistical Software tools SASused to be the best historically, many years ago
- 2Statistical Software tools STATAComes from economics, really good to handle large dataset
- 3Statistical Software tools Ropen source, thus free and often used in statistics teaching
- 4Statistical Software tools SPSSgraphically-oriented, thus historically the most-easy-to-use
- 5Preparation for Data Analysis- Coding
- Editing
- IMPORTANT:
never overwrite the original dataset - 6CodingTo categorise the responses for analysis
- 7Editing(To secure a quality standard)
- Fix errors
- Find obviously incorrect responses
- Allow processing - 8Coding(in this context) describes the process though which survey answers, texts, etc. are transformed into processable numbers
--> the steps of the process are defined in the code book - 9In surveys, in particular when using survey tools, the code book is programmed into the survey itself- That is, first, you define the question and the answers, and you assign a value to each answer
- Second, when the survey is done, you export all questions and answers - 10Missing values- Answer that are not processable (incl. "just not there") need to be defined as so-called missing values
- The issue is that if one variable is missing for an observation, that observation cannot be included in any more complex analysis - 11Answer that are not processable (incl. "just not there") need to be defined as so-called missing values- This means that you are telling your statistics software that even if there may be a number, this number cannot be used
- For example, you might have assigned the value of "-10" to the "Don't know" option - but how should the software know to treat the "-10" differently from the "1" to "5" you assigned to the real answers? - 12The issue is that if one variable is missing for an observation, that observation cannot be included in any more complex analysis-"Observation" here means your unit of analysis, i.e. who or what the data is about (e.g. an individual completing the survey)
- It is the same as a line in your dataset: if one field contais a missing value or is simply empty, it may well be that you have to throw away the entire line - 13Addressing missing values- Imputation
- Scales and indices
- Pair-wise statistics
- Drop observations - 14ImputationImputation describes a statistical process by which we try to infer the answers that are missing from the other answers
Very complex!! - 15Scales and indicesFor scales consisting of multiple items, it may not be necessary that all items are present
for example, 2 out of 4 may be enough - 16Pair-wise statisticsPair-wise statistics (such as pair-wise correlations between variables) always take the maximum number of combinations possible
- 17Drop observationsFor more complex analysis, however, we need to take the least common denominator (kleinster gemeinsame Nenner)
For those variables we are interested in, we can only take those observations where all variables are present - 18Deskriptive StatistikHat zum Ziel, empirische Daten durch Tabellen, Kennzeichen und Grafiken übersichtlich darzustellen und zu ordnen!
Sie macht keine Aussagen zu einer über die untersuchten Fälle hinausgehenden Grundgesamtheit und keine Überprüfung von Hypothesen ermöglicht.
Sie verwendet keine stochastischen Modelle, so dass die getroffenen Aussagen nicht durch Fehlerwahrscheinlihckeiten abgesichert sind. - 19The "first-moment" statistics Central tendency-Characterise data in terms of its central point
- Mode
- Median
- Mean - 20Modethe value most frequently observed
- 21Medianthe central value in ordered observations
- 22Meanthe sum of the values of a set of observations devided by the number of observations (the "average")
- 23The "second-moment" statistics Dispersion (Verteilung/Streuung)- Characterise data in terms of its spread around its central point
(Captures the degree to which observations are different from the mean
- Variance
- Standard deviation
- Range - 24Variancethe mean of the sqare of the differences between each value and the mean
Streuungsmaß, beschreibt die erwartete quadratische Abweichung der Zufallsvariablen von ihrem Erwartungswert - 25Standard deviationthe square root of the variance
- 26RangeDifference between minimum and maximum
- 27Standard errorTries to capture the degree to which the estimated value of a variable is different from its true value
Is thus a measure of the precision of an estimation (and not descriptive)
The standard error will be needed for statistical testing
SE(x) = s / Wurzel(n)
s= sample standard deviation of the respective variable - 28Looking at the formula, however, we can see two things already- A higher standard deviation increases the standard error
- A higher number of observations decreases the standard error - 29HintThus, any measurement can be made more precise by increasing the number of observations
- 30Data preparation as the necessary first step- Coding
- Editing - 31Codingrule book of operationalization
How to get your variables into your tool of preference - 32EditingData is never "clean"
Errors may have to be addressed manually
data can be processed further (generating new variables, factor analysis, etc.) - 33Hint!!Never overwrite the original data
Ideally, work with scripts so that you can reproduce every step of editing at any point in time - 34Descriptive statistics to provide information about your data- Core reports are usually the N, min, max, mean, median, std. dev.
- Correlations, correlation tables allow for a first look for patterns in data - 35Why do we need experimental research?- Descriptive studies allow us to describe how people think, feel, and behave
- Correlational studies allow us to see whether certain variables are related to one another
- But neither of these approaches allows us to know about the causes of behavior
- An experiment must be conducted to see whether or not one variable may be causing another variable - 36What is experimental research? Experiments are research designs in which the researcher..- manipulates one or more variables (uV)
- measures any change in other variables (aV)
- while controlling for the influence of extraneous variables (control variables) - 37The experimental setting RQ: What is the effect of money on helpfullnessLaboratory experiment in which helpfulness was operationalized as monetary donations
- 38The psychological consequences of money Participants were randomly assigned to one of two "conditions" (Groups)Control condition with a "neutral" task
experimental condition with a "money" task - 39Aufbau des Experiments1. Personen in unterschiedliche Gruppen
2. Aufgabenstellung und Bearbeitung
3. Allgemeine Befragung (non-related questionnaire)
Für beide Gruppen gleich, nur eine Füllaufgabe
4. Verabschiedung
5. Bevor Forscher Labor verlassen, wird darauf hingewiesn, dass man noch Spenden kann - 40Ergebnis des Experiments - Wer spendet mehr?1. Control condition
$1,34 (+- 1,02)
2. Experimental condition
$ 0,77 (+- 0,74) - 41Basic ingredients for designing experiments1) Independent variables
2) Dependent variables - 42Interdependent variablesin an experiment, the researcher manipulates one or more independent variables
an independent variable must have two or more levels (different values of the independent variable) - 43Hint Independent variablesA difference in levels is enough
Methodologically, no group has to be a "pure" control group - 44What is the independent variable in our money experiment?Exposure to money
- 45What are the levels at which the variable is manipulated?exposure (Enthüllung, Aufdeckung) to money vs. no exposure to money
- 46Dependent variablesdependent variable - the response being measured in a study, typically a measure of participants' thoughts, feelings, behavior, or physiological reactions
- 47What is the dependent variable in our money experiment?Helping behavior, operationalized as monetary donations
- 48Don't forget You can administer a survey at any time during an experiment- Get baseline variables (to control for confounding)
- Get further variables (usually additional independent variables)
- Manipulation checks
--> you can also use a survey to administer an experiment - 49Don't forget DebriefingIn particular in an experiment that uses deception, people need to be told what was done and why it was done after the experiment
Problem:
you still need to ensure that your other participants do not get confounded by learning about your experimental conditions - 50Five essential properties of an experiment Besides the three basic ingredients, there are five futher essential properties that characterize a well-designed experiment1) Control
2) Randomization
3) Systematic manipulation
4) Descriptiveness
5) Reproducibility - 51ControlIn order to assess the effect of the independent variable on the dependent variable, the researcher needs to control for the influence of extraneous (potentially confounding) variables
- 52Confounding variablean extraneous variable whose presence affects the variables being studied so that the results do not reflect the actual relationship between the variables under investigation
Example:
Control for membership in student associations, as members may be more likely to donate than non-members - 53How to eliminate the effects of extraneous variables?- Randomized allocation of participants to groups
- Statistical control of extraneous variables
(e.g., including extraneous variables such as day time in data analysis)
- Holding other variables constant (e.g., same instructions and questions for all groups - except for the experimental manipulation) - 54RandomizationRandom assignment
Participants are placed in conditions in such a way that every participant has an equal probability of being in any experimental condition
By using random assignment, the researcher can be confident that the groups are roughly equivalent at the beginning of the study - 55Randomization ExampleToss a coin for assigning participants to conditions
- 56Systematic manipulationThe researcher needs to purposely manipulate (vary) the independent variable in a systematic manner
- 57Many studies fail because the independent variable is not manipulated successfullyThis can be checked in a seperate pilot test (that solely focuses on the manipulation, and has nothing to do with the experiment)
This should also be controlled by a manipulation check, i.e. questions that check whether the IV has been sucessfully manipulated - 58DescriptivenessThe researcher needs to be able to thoroughly(gründlich/völlig) describe what exactly was done in the experiment, including...
- 59The researcher needs to be able to thoroughly(gründlich/völlig) describe what exactly was done in the experiment, including...- How was the uV measured?
- How was the aV assessed?
- How were participants assigned to groups?
- What exactly were participants requested to do? - 60ReproducibilityOther researchers need to be able to repeat(=replicate) the experiment based on the descriptions of the researcher who orginally conducted it
When repeated, the experiment should reveal the same result!! - 61When repeated, the experiment should reveal the same result!! What if not?Try to identify the cause why the result could not be replicated
Potential source of new knowledge - 62Threats to internal validity- Failed random assignment
- Different drop-out rates
- Extraneous events
- Experimenter expectancy - 63Failed random assignmentgroups are not
equivalent in the beginning (e.g., more female
participants in the experimental group, but more
male participants in the control group) - 64Different drop-out ratesparticipants drop out of
experimental/control conditions at different rates
(e.g., if the control task is “boring” but the experimental
task is “interesting”, more participants may drop out of
the control condition) - 65Extraneous eventsoccurring outside of the research
setting that have an effect on participants’ responses
(e.g., noise from outside) - 66Experimenter expectancymay be a serious threat
to internal validity
This effect occurs when a researcher’s expectations
about the outcome of a study influences participants’
reactions or the experimenter’s evaluations thereof - 67Double-blind procedureso avoid this, researchers use double-blind
procedures in which neither the participants nor
the experimenter who interacts with them know
which condition the participant is in
– Also important for ethical issues, think medical trials - 68The trade-off between internal and external validityThe more tightly controlled an
experiment, the stronger its
internal validity (and the more
accurate conclusions one can
draw about the effects of the
independent variable) - 69The trade-off between internal and external validityHowever, a high degree of
experimental control makes the
experiment more unique and
less like other settings, thereby
lowering external validity - 70Experiments as the "best research method" (?)A well-done experiment may be the only method that allows to find "true" causality
- 71Good experiments need to meet a series of conditionsManipulation of the uV, capturing of the aV, controlling for other effects
Clean setup:
control, randomization, systematic manipulation, descriptiveness, reproducibility - 72Experiments still have shortcomingsInternal vs. external validity
Remember McGrath!! - 73Number of independent variablesone factor at a time vs. Factorial
Factor-levels 1,2,...n vs. AxB-Design
AxB-Design:
Combination of factor-levels of A and B produces condition - 74Number of dependent variablesUnivariate vs. Multivariate
One DV vs. Multiple DVs - 75How is measurement repeatedBetween subject vs. Within-subject
- 76Note:Any combination of the above three dimensions is possible
- 77One factor vs. factorial designsRQ: What is the effect of room temperature on exam score?
- 78One factor option- "Temperature" with 2 options (e.g., 20°C, 30°C)
- 2 options could be tackled with simple t-test (more options would require a different test) - 79Factorial designs Changing the RQ for factorial DesignChanging the RQ for factorial Design
What is the (joint) effect of room temperature and the difficulty of the test on exam scores? - 80Factorial designs- "Temperature" as above
- In addition: "Difficulty" with 2 options (easy, hard)
- Results in a 2x2 between-subjects factorial design
- ANOVA to test the fields against each other - 81Between-subjects-designpeople are assigned to groups, comparisons are made across groups
- 82Within-subjects-designEach person is tested multiple times
- 83Experimental economics Guidelines- Incentivize subjects with real monetary payoffs
- Publish full experimental instructions
- Do not use deception
- Avoid introducing specific, concrete context - 84Experimental economics Contexts- Markets
- Coodination
- Learning
- Social preferences - 85Conjoint (gemeinsame) experiments Guidelines- Dissect test object into discrete dimensions (e.g. laptop: price, speed, weight) thought to be essential to utility of user etc.
- Define values for each dimension
- Derive reduced set of "cards" to capture values on each dimension (use software to help you)
- Ask people to rank cards or compare in pairs/groups - 86Conjoint (gemeinsame) experiments Context- Investment
- Purchasing
- Evaluation of teams - 87Natural experiment- One group of subjects is exogenously treated by a chance event, or at least something outside their control; another, similar group, is not
--> Natural disasters, "acts of God"
--> Change in laws
--> Lotteries, chance events
- Guidelines: just find it - 88Quasi experiment- Similar to natural experiment, but variable does not "occur naturally", but is selected upon by the researcher
--> children born into different types of households
--> Students assigned to different classes (if not by lottery)
- Guidelines: ibid - 89The standard paper/thesis/... structure...1) Introduction
2) Theory
3) Data and Methods
4) Results/Findings
5) Discussion/Implications
Informações
- Descrição
- ERM FlashCards sobre ERM 10-12, criado por Yannik Prinz em 19-02-2017.
- Criado por
- Yannik Prinz
- Visualizações
- 6
- Avaliação
Mais sobre Descriptive Statistics and Data Presentation