Please enable JavaScript.
Coggle requires JavaScript to display documents.
Chapters 11-13 - Mary Lisa Rider - Coggle Diagram
Chapters 11-13 - Mary Lisa Rider
Chapter 13 Qualitative Analysis
Qualitative analysis overview
Focus on sense making not prediction
Works with text data like interviews and observations
Depends on researcher judgment and context knowledge
Creative and investigative mindset needed
Emphasis on participant in context and ethical stance
Grounded theory
Inductive theory building from data
Theory is grounded in observed empirical evidence
Researcher suspends prior theoretical bias
Developed by Glaser and Strauss
Uses constant comparative method
Grounded theory steps
Open coding
Line by line reading of raw text
Identifying discrete events and ideas
Creating concepts in vivo codes
Linking concepts to text segments
Noting characteristics such as level or intensity
Categorisation
Grouping similar concepts into categories
Categories are broader and more general
Subcategories combine into higher order categories
Categories can evolve into constructs
Properties and dimensions are identified
Example property speed of communication media
Dimension fast medium slow
Axial coding
Connecting categories and subcategories
Building causal relationships and hypotheses
Identifying conditions actions interactions and consequences
Explaining why a phenomenon occurs
Explaining under what conditions and with what outcomes
Selective coding
Identifying central core category
Relating all other categories to core category
Selective sampling of new data to refine theory
Focusing analysis around central construct
Continuing to watch for new relevant categories
Theoretical saturation
When new data adds no new categories
When core categories and relationships stabilise
Signals that theory has been fully developed from data
Constant comparison process
Comparing incidents assigned to each category
Refining categories and their properties
Delimiting theory by focusing on core concepts
Writing theory through explanatory tools
Writing tools in grounded theory
Storylining
Building a narrative of the phenomenon
Using categories and relationships to tell a coherent story
Memoing
Writing analytic memos during coding
Capturing evolving ideas and theoretical insights
Reusing memos to detect patterns and relationships
Concept mapping
Graphical representation of concepts and links
Using boxes and arrows to show structure
Rearranging maps as understanding deepens
Content analysis
Systematic study of text content
Who says what to whom and why
Examines extent and effect of messages
Can be quantitative or qualitative
Steps in content analysis
Sampling relevant texts
Newspaper stories
Financial reports
Blog posts
Online reviews
Unitising
Dividing text into sections or chunks
Units such as assumptions barriers enablers effects
Coding
Applying concepts or themes to each unit
Using a coding scheme based on research goals
Analysis
Examining frequency of themes
Examining context of themes
Exploring relationships between themes
Sentiment analysis as content analysis
Focus on opinions and attitudes
Classifying messages as positive negative or neutral
Example online comments about political candidates
Example online product or service reviews
Can be manual or automated with text analytics tools
Spiral model of content analysis Schilling
Level one convert recordings to transcripts
Level two condense raw text into shorter protocols
Level three build preliminary category system
Level four apply categories to produce coded protocols
Level five analyse coded protocols to interpret phenomenon
Limitations of content analysis
Depends on availability of text sources
Sampling bias if sources are not representative
May miss non textual aspects of experience
May oversimplify complex meanings
Hermeneutic analysis
Special type of content analysis
Focus on interpreting subjective meaning of text
Strong attention to sociohistoric context
Uses hermeneutic circle
Moving between parts of text and whole context
Deepening understanding through repeated reading
Assumes text narrates author experience in context
Hermeneutics broader view
Origin in interpreting religious legal and literary texts
Traditional hermeneutics example biblical interpretation
Philosophical hermeneutics Heidegger
Emphasis on experiencing reality not just observing
Texts let readers relive author experiences
Modern hermeneutics Gadamer and others
Includes all forms of communication
Considers presuppositions and language structure
Uses semiotics symbols metaphors and analogies
Exegesis contrasted with hermeneutics
Exegesis is critical explanation of text itself
Often applied to religious texts
Focuses on what is written not full context
Sometimes mistakenly used as synonym for hermeneutics
Qualitative analysis software
Examples ATLAS ti NVivo QDA Miner
Can automate some coding processes
Organise search and sort large volumes of text
Require coding schema based on initial manual review
Coding schema can be hierarchical with higher order codes
Schema must be validated on sample texts
Risk of biased or incorrect schema
Software cannot fully interpret context sarcasm or metaphor
Human judgment still essential for meaningful interpretation
Chapter 14 Quantitative Analysis Descriptive Statistics
Overview of quantitative analysis
Works with numeric data
Uses statistical tools and software
Two main types descriptive and inferential
Chapter focus is descriptive statistics
Descriptive analysis
Describes variables and relationships
Summarises data in tables and charts
Helps understand what the sample looks like
Inferential analysis contrast
Tests hypotheses about populations
Uses sample data to draw conclusions
Covered in Chapter 15 not here
Data preparation overview
Needed before running any statistics
Convert raw data into numeric form
Make data machine readable for software
Data coding
Assign numbers to responses
Use a codebook to define everything
Variable names
Response scales and formats
Coding rules for each value
Examples
Likert scale from strongly disagree to strongly agree coded 1 to 7
Nominal categories like industry coded as 1 2 3 etc
Ratio data such as age and income entered as given
Aggregating data
Summing checklist items to create a total score
Turning many observed answers into a single measure
Data entry
Enter coded data into spreadsheet or database
Software options include Excel SPSS SAS
One row per case or respondent
One column per item or variable
Check for accuracy with spot checks
Watch for bad data patterns
Example same answer to every item including reverse coded ones
Problematic cases can be excluded from analysis
Missing values
Common in real datasets
Reasons
Ambiguous questions
Sensitive topics
Handling missing values in software
Some treat blanks as missing automatically
Others need a special missing code such as minus one or nine nine nine
Listwise deletion
Removes any case with a missing value
Can reduce sample size
Makes it harder to detect effects
Imputation
Replacing missing values with estimated values
For multi item scales can use person’s average on other items
For single item scales can use group average
Assumes missing data are random
Advanced imputation methods
Maximum likelihood procedures
Multiple imputation methods
Available in SPSS and SAS
Data transformation
Reverse coding
Items that are worded opposite of construct
Example subtracting value from maximum plus one
Creating scale measures
Summing or averaging items into single index
Creating weighted indices when items have different importance
Collapsing categories
Grouping detailed values into broader ranges
Example income brackets instead of exact income
Univariate analysis
Focus on single variable at a time
Used to explore basic properties of each variable
Key tools are frequency distribution central tendency dispersion
Frequency distribution
Shows how often each value occurs
Can use counts or percentages
Example religiosity categories from never to several times per week
Often displayed as tables
Can be shown as bar chart
Horizontal axis categories or values
Vertical axis frequency or percentage
With large samples distributions may approach normal curve
Normal distribution
Bell shaped curve
Most observations cluster near center
Fewer at extremes
Basis for many statistical assumptions
Central tendency
Represents center of distribution
Mean
Arithmetic average
Sum of values divided by number of values
Median
Middle value after sorting
If even number of values average of two middle values
Mode
Most frequently occurring value
All are statistics estimated from sample
Dispersion
Shows spread of values around center
Range
Highest value minus lowest value
Sensitive to outliers
Standard deviation
Average distance of values from mean
Uses formula based on each value’s deviation
Variance
Square of standard deviation
In normal distribution
About 68 percent of cases within one standard deviation
About 95 percent within two standard deviations
About 99 point 7 percent within three standard deviations
Bivariate analysis
Examines relationship between two variables
Most common tool is correlation
Correlation
Measures strength and direction of linear relationship
Value ranges from minus one to plus one
Positive correlation both variables increase together
Negative correlation one increases while other decreases
Zero correlation no linear pattern
Can be visualised with scatterplot
Significance testing for correlation
Null hypothesis r equals zero no relationship
Alternative hypothesis r not equal zero or r greater than zero
Can use one tailed or two tailed tests
Uses t tests and critical values of r
P value and significance level
P value probability that result is due to chance
Significance level alpha usually zero point zero five
If p less than alpha reject null hypothesis
If p greater than alpha cannot reject null hypothesis
Interpretation is always probabilistic
Degrees of freedom in correlation
Equal to n minus two
Used to look up critical values in tables
Correlation matrix
Table of correlations between many variables
Variables listed in first row and first column
Diagonal values equal one correlation with itself
Upper and lower triangles mirror each other
Often only lower triangle reported
Pearson correlations when using interval scale measures
Cross tabulation basics
Also called crosstab or contingency table
Used for relationships between categorical variables
Shows joint distribution example gender by grade
Helps reveal patterns in proportions
Marginal totals show row and column sums
Chi square test
Tests whether observed pattern could occur by chance
Uses expected counts based on marginal totals
Compares observed and expected counts in each cell
Computes chi square statistic
Uses degrees of freedom equal to rows minus one times columns minus one
If computed value exceeds critical value pattern is significant
Chapter 15 Quantitative Analysis Inferential Statistics
Overview of inferential statistics
Used to test hypotheses
Move from sample data to population conclusions
Always probabilistic never certain
Supported by software like SPSS and SAS
Popper and falsification
Theories cannot be proven only disproven
Alternative hypothesis cannot be directly proved
We look for evidence to reject null hypothesis
Absence of contrary evidence does not guarantee truth
Null and alternative hypotheses
Null hypothesis represents no effect or no relationship
Alternative hypothesis represents effect or relationship
Statistical tests focus on rejecting null hypothesis
Rejection provides indirect support for alternative
Extraneous variables problem
Dependent variable influenced by many uncontrolled factors
Cannot measure or control everything
Sample relationships may not hold in population
Reinforces need for probabilistic reasoning
Fisher and significance testing
Established five percent rule for significance
P value probability result is due to chance
Significance level alpha is maximum acceptable risk
Common alpha is zero point zero five
If p less than or equal to alpha result is statistically significant
Type I error is rejecting null when it is actually true
Sampling distribution
Theoretical distribution of statistics from infinite samples
Each sample estimate has some error
Variation in sample statistics captured by distribution
Standard error
Measure of sampling error in statistic
Smaller standard error means better estimate of population value
Depends on sample size and variability
Confidence interval
Range around sample estimate with specified confidence level
Example ninety five percent interval about plus or minus two standard deviations
Means we expect population value to fall in that range most of the time
Combined with p value to judge reliability of results
General linear model GLM
Family of linear statistical models
Represents relationships as straight line equations
Basic two variable model
Outcome y predictor x
Equation y equals beta zero plus beta one x plus error
Beta coefficients are regression parameters
Error term captures unexplained variation
Multiple predictors
Y explained by several x variables
Equation includes beta terms for each predictor
Covariates
Control variables included to account for systematic effects
Allow clearer estimation of key predictors
Dummy variables
Nominal variables coded as zero or one
Used for groups like gender or industry
Types of GLM based methods
Simple regression
One predictor one outcome
Estimates slope and intercept
Multiple regression
Several predictors one outcome
Estimates contribution of each predictor
ANOVA analysis of variance
Categorical predictor for group comparisons
Equivalent to regression with dummy variables
ANCOVA analysis of covariance
Combines ANOVA and regression
Includes covariates to adjust group comparisons
MANOVA multivariate ANOVA
Multiple outcomes with ANOVA framework
MANCOVA multivariate ANCOVA
Multiple outcomes with covariates included
Structural equation modelling
System of linked regression equations
Allows complex directional relationships
Requires careful model specification
Model specification
Choosing which variables and paths to include
Should be based on theory not just data fit
Data used to validate not define models
Two group comparison
Common inferential design
Example treatment versus control groups
Predictor is dummy variable zero or one
Outcome is usually ratio scale outcome such as test score
Analysed using t test and one way ANOVA
Student’s t test
Introduced by William Sealy Gosset
Compares means of two groups
Can be one tailed or two tailed
Test statistic t equals difference in sample means divided by standard error of difference
Standard error based on variances and sample sizes
Degrees of freedom approximately n1 plus n2 minus two
P value from t distribution compared to alpha
If p less than alpha means are significantly different
Effect size in two group comparison
Shows magnitude of treatment effect
Can be estimated with regression coefficient beta one
When predictor is dummy effect size equals difference in group means
Helps interpret practical importance beyond significance
Factorial designs
Examine two or more factors at once
Example curriculum type and instructional time
Main effects
Independent effect of each factor
Interaction effect
When effect of one factor depends on level of another
GLM for factorial
Includes terms for main effects and interaction term
Interpretation rules
If interaction is significant do not interpret main effects alone
Covariates can be added as additional predictors
Other quantitative analyses
Factor analysis
Data reduction technique
Groups related items into latent factors
Used to assess scale validity convergent and discriminant
Discriminant analysis
Classification technique
Predicts membership in nominal categories
Similar to regression but outcome is categorical
Used in marketing for customer or product classification
Logistic regression
Outcome is binary zero or one
Uses logistic curve to model probability
Common in medical and social science research
Effect size interpreted with odds ratios
Probit regression
Outcome between zero and one or binary
Assumes normal distribution of underlying variable
Popular in finance insurance and risk modelling
Coefficients often similar to logistic regression
Path analysis
Multivariate technique for directional relationships
Uses multiple regression equations in one system
Useful for testing complex theoretical models
Time series analysis
Analyses data that change over time
Examples stock prices crime rates economic indicators
Deals with autocorrelation across time points
Used in econometrics and signal processing