reading 4 Flashcards

1
Q

Two main sets of methodologies is changing the way organizations do business in the current digital age:

A

big data analytics: data mining, machine learning and other statistical techniques allow practitioners to handle and analyze huge sets of data with a reasonable effort

Business field experiemnts: studies conducted outside of the lab by means of ease-to-use software allow managers to reliably answer causality questions at reasonable costs

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

When should I consider setting up a field experiment to answer my research or business questions

A

field experiments are one of the most reliable ways to test a theory or to prove that a business action results in a desired outcome

Findings from field experiments have direct implications for business operations. In the language of experimentation, we say that they generalize well and have high external validity. On the other hand, lab experiments are acknowledged to have higher internal validity

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

True experiments (either in the lab or the field) show three identifiable aspects

A

1) the responses of experimental subjects assigned to receive one treatment are compared to the responses of subjects assigned to another treatment (often a control gorup which receives some type of baseline treatment that is essentially no treatment or the state of the art condition. In the case of multivariate experiments, there are several treatment groups, which are all compared among each other

2) The assignment of subjects to each group is done through a randomization device such as a coin flip, a dice roll or algorithm
3) the manipulation of the treatment is under the control of an experimental researcher

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Natural experiments

A

share attribute 1 and partially atrtribute 2 of t rue experiments since assignment is random or as if random. However in such cases, data comes from naturally occurring phenomena and therefore the manipulation of treatment variables is not generally under the reserachers control

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Lab experiments

A

Participants are tested in an environment which is created by the researcher and which thus differs from reality. This unreal environment allows the experimenter to control other potential influences on the response but has the main drawback of making the respondent feel observed, which can lead to serveral kinds of response bias

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Field experiments

A

in which the setting is an everyday life situation, often the exact same setting where the findings from the experiment will be deployed. IN most field experiments, participants are not even conscious of taking part in an experiment eliminating the risk of incurring a response bias

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Key features of field experiments

A

authenticity of treatments, representativeness of participants, real world context, and relevant outcome measures

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Causal effect:

A

the difference between tow potential outcomes, one in which a subject receives the treatment and the other in whic hthe subject does not receive the treatment

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Experiments both in the lab and field provide unbiased estimates of the ATE when teh following assumptions are met:

A

1) random assignment: treatments are allocated such that all units have an equal probability between 0 and 1 of being assigned to the treatment group

2) excludability: the treatment must be defined clearly s otaht one can assess whether subjects are exposed to the intended treatment or to something else

3) noninterference: no matter which subjects the random assignment allocates to treatment or control, a given subjects potential outcomes remain the saem

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Internal validity:

A

refers tot he extent to which we can say the observed effect in our study was caused by our treatment

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Selection bias

A

when assignment to treatment is not random and certain types of people are more likely to receive one of the treatments, in other words, the experimental groups systematicall differ form each other

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Differentia lattrition:

A

when certain types of subjects drop out of one of the treatments. it implies that certain types of participants leave during the run of the experiment or do not take part in the final measurement

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Time effects

A

when treatments are administered at two different times, outside events, learningor other changes are confounded with the treatment

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Confounding variables:

A

when other vairables are correlated with the treatment and have an effect on the outcome, a cause effect relationship between the confounder and the dependent variable can be mistakenly assumed to be a causal effect of the treatment

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Noncompliance

A

subjects are assigned to the experiment do not get the specified treatment. THis can happend because of individuals voluntary decision to use a different treatment than the one they were assigned, because they do not like it or they think another treatment would be better

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Diffusion of the treatment across groups

A

subjects assigned to one treatment find out about the other treatment

17
Q

demand effects

A

participants guess the hypothesis of the experiment and try to cooperate by exhibiting behaviour that confirms the hypothesis

18
Q

Experimenter bias

A

makes subjective measurements and inadvertently favors the hypothesis in those measurements

19
Q

Hawthorne effect

A

it is also possible that individuals being part of an experiment and being monitored change their behaviour due to the attention they are receiving from researchers rather than because of the manipulation o fthe independent variables

20
Q

Ambiguous temporal precendence

A

in some experiments it can be unclear whether the treatment was administered before or after the effect was measured. For instance, if purchases and promotional emails are tracked at a daily level, it can be difficult to discern if a customer who received an email on a particular day and also maede a purchase the same day received the email before she made the purchase

21
Q

Simple random sample

A

takes a random draw from the target population using for instance a coin flip or a dice roll. This gives to each subject an identical probability of entering the sample, ensuring that the sample will be representative of the target population

22
Q

cluster sample

A

when it is easy to measure groups or clusters of subjects, randomly sample from among the clusters

23
Q

stratified sample

A

use a procedure to make sure that the sample contains different types of subjectscon

24
Q

convenience sample

A

sample in some way that is easy for the researcher, e.g. an academic might conduct the experiment with students or a company might conduct the experiment using store locations that are nearby

25
Q

interaction effect

A

when the combined effect of two treatments is better than the sum of the individual effects, there is an interaction effect. Detecting interactions is the main reason why companies use multivariate tests. In addition multivariate tests can reduce required sample sizes and increase the amount that can be learned in the time frame of a single test.

26
Q

An experiment requires two stages

A

a design state before the experiment is implemented, including, for instance, the selection of the initiative as well as the number and randomization of groups, and an examination stage after the experiment took place to analyze the data which had been generated

27
Q

The term A/B testing is often used when

A

referring to field experiments in online environments meaning that for example, one randomly selected group of customers is presented with packaging design A and another gorup is presented with design B for an otherwise identical product

28
Q

Control over treatment

A

describes the researchers ability to design and have full control over the type of treatment and the selection of subjects and circumstances

29
Q

Randomness of assignment

A

describes whether the experimental design classifies subjects randomly (or not) to the treatment and the control group. Here we differentiate between random and nonrandom assignment

30
Q

The differences in differences methodology

A

exploits two different comparisons within one design: An intertemporal comparison and a cross sectional comparison. The first comparison is the intertemporal difference within the treatment group; thus, the difference in observation before and after the treatment only for the gorup which has been affected by the manipulation

The second difference constitutes the variation between the treatment and the control group. here, observable outcomes of treatment and control group are compared for one or more points in time only after the intervention, that is, a cross sectional difference is exploited. In combining both differences, intertemporal and cross sectional, the differences in differences methodology allows to draw causal inferences as long as it fulfills specific conditions

31
Q

Staggered designs

A

here researchers either design their field experiment in a way, or exploit the fact that treatment does not occur to all treated individuals or groups at the same time but is introduced step by step. Staggered designs are often observed when regulations or legal directives are adopted or firms subsequently roll out a new policy over subjects such as business units, facilities, employee groups or customersT

32
Q

Three more assumptions with relevance to the differences in differences method and provided reasoning to which extent the sepcific research design meets those assumptions

A

The treatment status of one unit should not interfere with another units outcome. FOr example customer As status (qualified for free shipping) should not affect the amount of money spent by another customer B.
Neither subject in the treatment or in the control group anticipates the intervention or is affected by the treatment prior to the intervention. This assumption might be violated in cases where subjects can voluntarily choose to become part of a group, that is, a customer may decide to purchase an additional item to cross the threshold of the free shipping minimum order amount
Finally, the difference in differences method implicitly assumes perfect compliance. This means all individuals in the treatment group are in fact treated. In case of a violation, the method estimates (just) an intention of treat effect. THis intention of treat effect is more conservative than the treatment on the treated estimation because it is based on a smaller variance within the data set making it harder to detect a statistically significant effect between treatment and control group. For free shipping threshold it may occur that certain customers may either simply not recognize that free shipping is available or at which threshold the would qualify for the free shipping