module 8 Flashcards
Internal archival data
Data that are generated within the company that conducts the research
E.g. A researcher employed by unilever would like to understand to what extend Unilevers advertising expenditures for their brands are related to brand sales. To answer this question, you use unilevers sales and advertising data for 400+ brands - data that are available within the company
External archival data
Data that are generated by sources outside a company, and that are not exclusively available to the company but can be used by anyone. Some archival data are publicly available. This means that they are available for free for the entire world to use. Other archival data are commercially available. This means thaat one has to pay to use them
Publicly available data:
Government publications
World bank (world development indicators)
OECD
Annual reports
Commercially available data sets:
Compustat, ORBIS (databases with accounting data)
Datastream (stock price data)
SDC (alliance data)
Execucomp (CEO data, incl. executive compensation)
Nielsen data (consumer and retailer panel data)
When is archival research a suitable research strategy
1) When you want to learn from past successes/failures in the industry
2) When you want to know whether/how an effect changes over time
3) When you want to know whether/how an effect is different across countries
4)When you want to examine socially sensitive phenomena unobtrusively
strenght 1: archival research
Tapping into industry wisdom:
Learn from past successes and failures in the industry when you cannot rely on your own experience
Strength 2 archival research
Examining effects across time
Examine whether an effect (IV –> DV) changes over time
strength 3 archival data
Examining effects across countries
Examine whether an effect (IV –> DV) is different across countries
strength 4 archival research
Examining socially-sensitive phenomena
Examine socially-sensitive phenomena unobtrusively, without collecting data directly from evaluees.
Minimizes the opportunity of distored responses
Piecing together archival data
Often you need to piece together multiple archival data sets into one new data set that you can subsequently use to address your research questions
What is important when pieceing together archival data
Your unit of analysis has to correspond
Unlike other dat sources, archival data can be (and typically are ) collected over time. The data are longitudinal (rather than cross-sectional, which means at one point in time) in nature. This makes them ideal if the nature of your research is to examine changes over time
Three sources of measurement unreliability may lague any type of research, but are particularly common in archival research. These are:
1) missing data
2) inaccurately recorded data
3) Fake data
Missing data in cross-sectional data sets
Solutions:
Listwise deletion: delete if there is missing info
BUT: check user manual! if missing = close to 0 then recode missing to 0
Mean substitution:
Replace missing value for observations I and variable with average value on variable J for all other observations
Missing data in longitudinal data sets
Solutions
Interpolation
E.g. average of year before and year after
Inaccurately recorded data
Solutions
Inaccuracies that turn up as extreme data points (outliers)
Inaccuracies that are not extreme
Solution: Remove observation: run analyses with and without observation
Trim/truncate (in large data sets)
Remove a fraction of observations
e.g. 1% most extreme observations
Fake observations Solutions
Be critical :
Where? when? why?
Composite measures
Many of the measures used in archival research are single-item measures that capture very concrete constructs. Think of measures such as stock returns, shareholder value, firm profitability, sales, or R&D expenditures.
However in a few instances multi-item measures can be desirable in an archival research study. This is the case when you need to measure a less concrete construct for which multiple potential measures exist.
For example, suppose you would like to investigate if firms become more profitable when they grow larger. How are going to measure the variable firm size? Multiple potential measures exist (and all of them have been used in prior research):
The firm’s sales (larger firms have more sales)
The firm’s number of employees (larger firms employ more people)
The market value of the firm (larger firms have a higher market value)
Which one of these three measures will you use to measure the variable firm size? If there is no good reason why one measure would be better than another, you can of course arbitrarily pick one of the three measures and continue with your research. Alternatively, you can create a composite measure that combines all three measures, provided that the three measures have a high Cronbach’s alpha (that is, they should be interrelated as you consider them to be indicators of the same underlying construct).
Combining multiple archival indicators into a single measure
1) standardize each indicator
2) average the standardized indicators
Measurement validity
Pertains to whether your measurement instruments capture what they are supposed to measure.
Considerable conceptual overlap: valid measure
Little conceptual overlap: questionable proxy
Questionable proxy example:
Construct: Firm innovativeness
measure: R&D spending
Imput doesnt mean results, you can spend a lot of money on R&D and not be innovative
Good measure would be output of new products
How to validate that an archival measure is a good measure rather than a bad proxy
Provide precedence, but
Provide sound logic to support that considerable conceptual overlap exists between construct and measure
Show that there is a substantial correlation between your proxy and a valid survey measure for a small subsample of data
How to validate that an archival measure is a good measure rather than a bad prody? (cont’d)
Provide evidence of substancial correlations (r>.3) with related constructs (nomological validity)
Internal validity
Internal validity is the extent to which a sstudy can rule out alternative explanations. To establish a valid relationship between variables, a researcher needs to limit the influence of extraneous/confounding variables. The less change there is for confounding in a study, the higher the internal validity, and the more confident you can be in the studys findings
How can you improve the internal validity of archival research
By including control variables in your analysis
by including control variables you can filter out or isolate the control variables effects from the relationship between the variables of interest
External validity (generalizability)
always read the documents explaining the methodology behind archival database
