content analysis quantitative evaluation of texts
TRANSCRIPT
Content analysis
Quantitative evaluation of texts
Questions about content
When we talk about “cop shows” or “news” or “sports” we think about certain kinds of content. Usually, we perceive certain regularities in the content and notice when a single ‘text’ or ‘artifact’ deviates from those expectations.
Sometimes we think certain regularities exist, while others dispute our beliefs.
The need for careful analysis
Because our own hunches and expectations can be in error, and much of our understanding of the effects, values, and role of telecommunications is dependent upon the nature of the content of television, radio, film, videogames, etc. it is often necessary to more carefully analyzed that content.
Text analyses
Many ways of evaluating content/texts are available. We call the entire group of methods “text analysis.”
The most heavily quantitative form of text analysis is “content analysis.”
Definition
Content analysis: A research technique for making inferences by systematically and objectively measuring specified characteristics of a text.
Content analysis in telecommunications research
Content analysis is probably the most common form of research found in scholarly study of telecommunications
It demands the least money and resources The downside is that many consider it “an
easy publication” and produce very low-quality work
Goals of text analysis To explain the nature of communication.
Describe the content, structure, and functions of the messages contained in texts.
What does the text mean? How does it achieve that meaning? To describe how communication is related to other
variables. Input variables – Outcome variables For example: How does a corporate takeover affect television
news coverage? To evaluate texts by using a set of standards or
criteria. Must establish a set of standards against which the
communication can be compared. Example: Is the text too hard to read for 12-year-olds?
Types of texts
Most any fixed symbolic whole—a story, a textbook, a church, a transcribed conversation, a website, and on and on can be considered a ‘text’. Sometimes a whole series of stories (Star Trek, season 2) may be considered a ‘text’.
Acquiring texts
Listen to conversations in naturalistic settings
Conversations produced in a lab Visit rooms of teenage girls Literary or historical sources (novels or
films) Record shows off the air Visit or mirror websites
Procedures
Select the text(s) to be analyzed Determine the recording units Develop content categories Train observers to code units into categories Carry out the coding while monitoring for quality Analyze the data
Sampling in content analysis
Population: totality of texts we want to say something about
This is often more difficult than it seems All issues of the Herald Leader over a period of a
year? All coverage of terrorism in the elite press?
We can analyze a census or we can sample The same sorts of sampling techniques used for
surveys can be applied here Random v. non-random sampling Many non-random samples chosen for theoretical as well as
convenience reasons
Sampling
Commonly multiple stages in sampling documents
Selecting communication sources Newsweek Prime Time television
Sampling documents Pick an issue, particular shows
Sampling within documents Front page v. all pages, etc.
Units of observation
Chosen first according to theory, then by convenience Articles Broadcasts Books Pictures Movies Letters Conversations
Recording units
Recording units are the actual ‘pieces’ of the observational units that are scored according to your category scheme
For example, if I were observing a single episode of NCIS, I might score every 5 minutes of the show for the presence or absence of humor. The 5-minute segment would be my recording unit.
Recording units
Single word or symbol May be too small—large number of data points generated
Theme Single assertion about some subject May have overlapping themes
Character Person or animal categorized rather than words or themes
Sentence or paragraph May have ambiguous or conflicted evidence of one or more categories
of content Item
Whole book, film, radio program Difficulty coding into single categories
Physical size measure Column inches Number of seconds
Coding categories
The category scheme is the set of dimensions you use to evaluate your recording units and the available options you have for scoring one recording unit on each dimension For example:
Recording unit: Scene Dimension: Emotionality of a scene Scoring options: High/Medium/Low
Coding categories
The coding categories must be carefully developed in order to see that when the actual data are generated, they answer your theoretic questions of the text I can’t tell you how many times I’ve had to
reject manuscripts because the coding scheme was not adequate to answer the theoretical questions posed in the literature review
Coding scheme
Conceptualization coding categories The code book provides the rules for
assigning a coding unit to one or another category It is an actual set of rules for assigning the
proper codes (scoring) to each coding unit
Coding rules
What are the rules for determining which category a given recording unit should be placed in? How do we know whether a given paragraph is
pro-Lunsford, neutral, or anti-Lunsford? This is a crucial part of the coding scheme.
A naïve coder who simply applies the rules should get the outcome the theorist/ researcher intended.
Good coding categories
Categories should be: Exhaustive Mutually exclusive Derived from a single classification principle Independent Adequate to answer the questions asked of
the data
Practice coding
In order to see that coders use the instrument as the researcher intended, the researcher holds practice sessions Related content, usually not from the actual
sample, is coded and the results discussed
Coding sheet example
Coding units
Coding categories
Length of song (secs)
# different words
Main topical focus
# instruments played
Vocal enhancement
Oops I did it again
We are the world
Stairway to heaven
Unchained melody
Coding reliability
To ensure that the coding scheme is reliable we have to test it
Coders score identical content The more often different coders produce the same
scores for the identical content, the more reliable the coding scheme is
Results are compared using statistical tests for reliability Cronbach’s alpha; Krippendorff’s alpha
A rule of thumb is that the coding scheme is reliable if alpha is at least .70
Reliability v. Validity v. Precision
The highest levels of reliability are usually found with very simple, extreme codes (true v. false; happy v. sad) but these simple codes often don’t provide the precision we want (clearly true, seemingly true, ambiguous, seemingly false, clearly false) and therefore reduce the value of the results—validity may suffer.
The researcher has to consider the tradeoff
Data analyses
Descriptive statistics are often used Percentages Mean Standard deviation May compare across texts
To test hypotheses, etc. Compare findings to some prediction
Relative percentages among categories, between sources on same categories
Correlations among categories, with predictor variables, with outcome variables E.g., goriness of violence with measures of audience
enjoyment