Please enable JavaScript.

Coggle requires JavaScript to display documents.

NUR4103 Applied Research Methods - Coggle Diagram

- - - - Types of interviews
        
        Individual
        
        Focus grp
        
        Generate data not possible through indiv interviews
        For discussion and debate
        Not feel safe/comfortable sharing alone
        6-10 per FGD, minimum 3 FGDs
        Prevent group speak --> intervene and encourage the rest
      - Structure of interview
        
        Semi-structured
        
        Interview guide prepared to allow for freedom and probe to elicit in-depth information
        
        Design interview guide
        
        Ensure interview questions align with the research questions
        
        Constructing an enquiry based conversation
        
        Receive feedback on interview guide
        
        Pilot the interview guide
        
        Step 2
        Balance between an inquiry and conversation
        Intro qns, transition qns, key qns, closing qns
        Intro: how was it when you were taking care of your child
        Transition qn: you were sharing this was stressful...
        KEY QN: How was support received ("I want to specifically ask you...)
        Closing qn: Is there any more you want to share?
        Generic, what the person is saying, key qns, closing
        
        Step 3:
        Receive feedback from experts to refine the question
        
        Step 1
        Questions based on lit review, experience, consultation
        Ensure questions align with research questions
        
        Step 4
        Pilot interview guide and make final revisions
        Simulate actual interview
        
        Interview process
        
        Initiation phase
        Build rapport
        Address participant, introduce self and explain role
        Formulate contract (length of session, confidentiality issues, compensation)
        Obtain consent to your interview plan (asking permission)
        
        Exploration phase
        Broad to specific
        Show genuine interest
        Exhibit non-judgemental
        Give feedback/clarifications
        Allow silence (get used to awkwardness)
        AVOID LEADING QN
        
        Preparation phase
        Anticipate, plan, engage in self-assessment, materials/equipment, environment, timing, safety
        
        Termination phase
        Summarise most important contents
        ASKING
        Is there anything further you'd like to tell me or ask me?
        "Is there anything else we havent covered that you would like to discuss before I examine you?
        Expressing appreciation
        
        Unstructured
        
        Fully structured
  - - - Focus on people's lived experiences and interpretations
      - Difference with descriptive?
        Way to collect data. Longer, and more indepth compared to a single interview
      - Descriptive vs Interpretive phenomelogy?
        Explaining of phenomena; describing.
        Interpretive: researcher introduce their own facts and understanding
      - Epistemology: no preconceived idea, no literature review
        Sample size: 5-25 participants
        Data collection: selecting information rich cases (purposive sampling), complete and rigorous aprticipation
        Data anaylsis: transform what is seen and heard into conceptual categories to understand the original experience
        Results: exhaustive description of experiences without distorting or losing the rich data
    - - Studying a case within real life context or setting
        Can be single (single case) or mutliple (multiple bounded system cases)
      - For rare conditions/exceptions
      - Data collection: presents an in-depth understanding of the case
        Data analysis: rich description of the case; themes/issues can be organized chronologically; anlysis across cases for similarities and differences
        Results: presented as a theoretical model; assertions of general lessons learned
    - - Method of describing a culture or society
        Not a study of people, but LEARNING FROM PEOPLE
      - Need to be part of culture so people in culture not sensitive being observed and act differently
      - Epistemology: no preconceived idea, no lit review
        Sample size: 30-50 interviews
        Data collection: researcher as instrument, fieldwork
        Data Analysis: ethnographic records, cultural inventory
        Results: write an ethnography, themes, concepts
    - - Naturalist/constructivist inquiry
        Describe a phenomenon/depict the participants in an accurate way
      - Epistemology: may/may not have preconceived idea, may or may not do literature review (learn from experts) the particpant
        Sample size: 15 smallest acceptable sample
        Data collection: record observations/shared information in details using audio, videos and notes; construct a complex picture of the participant and environment
        Data analysis: thematic or content
        Results: presented as themes
    - - Main purpose is to develop theory
        Develop theoretical connections between concepts/dimensions
      - Developed by Glaser and Strauss
        No lit review, preconceived idea
        Sample size: 20-30
        Data collection: unstructured interviews, observations, photos, news documents
        Data analysis: memos, concepts, theoretical connections
        Results: themes, concepts, theory
- - - - Block randomization is to ensure equal number of intervention and control in limited sample size
      - Randomization =/ random sampling
        Random sampling: drawing a sample from population
        Randomization: dividing the subjects into 2 (or more groups) where the subjects can be from random sampling or convenient sampling
        Allocation concealment: concealing information about which patient are to be assigned to an intervention and control group = to reduce selection bias
    - - Crossover design
        
        Involving exposing subject to more than one treatment group
        There is WASHOUT PERIOD before control and intervention swapped
        Influence of covariates reduced since both go through the same
        Useful for small samples
      - Posttest only group design
        
        No pre-test because:
        
        Impossible to get pre-test measurement
        
        Pretest sensitization problem: sensitized subjects so respond differently
        
        Limited resources in collecting a pre-test measurement
      - Pretest-postttest control group design with repeated measure
        
        Posttest more than 1
        Longtitudinal changes observed
      - Solomon four group design
        
        Combine pretest-posttest + posttest control group design
        To examine if pretest has effect
      - Pretest-posttest control group design
    - - Differences between RCT:
        
        Some type of comparison group (may or may not have control grp)
        
        Do not contain random assignment of the subjects (no randomization)
      - Useful for change of outcome over time
      - Quasi experimental designs
        
        Nonequivalent control group design
        
        Pretest-posttest comparison group design
        
        Similar to pretest-posttest control group design but no randomization involved
        -randomization not possible
        -easier to implement without randomization
        E.g.Evaluated a simulation-based teaching strategy for improving English on 101 nursing students. Y1 + Y3 students in control, Y2 and Y4 in intervention
        
        Posttest only comparison group design
        
        Subjects are not randomly assigned to the intervention/comparison group
        
        Similar to posttest only control grp. Difference: no randomization
        Used when pre-test impossible/difficult to conduct
        E.g. Evaluate intervention of baby program.. Assigned to grps based on residential address. Follow up survey after intervention
        
        Interrupted time series design
        
        Interrupted time series design
        
        Subjects may/may not be the same
        Provides an estimate of the causal effect of a discrete intervention
        E.g. 20 cents added to a policy change. Compare with post-intervention and pre-intervention (look at word doc for clearer image)
        
        Design without a control grp
        
        One grp posttest only design
        
        Assume low knowledge on subject
        E.g. SIDS in small town of Maryland with media intervention. Test knowledge afterwards with survey
        SARS outbreak, handhygiene intervention
        
        One grp pretest-posttest design
        
        No randomization, no control grp. Similar to the design on top but pretest present
        E.g. hand hygiene practices of health care workers using single observer before and after an educational program
    - - Case study
        
        Starts from identifying subjects with and without diseases
        Also used to investigate etiology
        2 groups: Case (subject with disease) and control (subject with no disease)
        Exposure: harmful (risk) factors vs protective factors
        E.g. smoking; compare number of ppl with lung cancer are smokers or non-smokers
        Usually want incident cases (newly diagnosed) instead of prevalent (people have the disease for some time)
        
        Compare exposure proportion
        Usually P(disease) > P (no disease)
        Odds ratio (P (disease) x (1-P (no disease))/(1-P (disease) x P (no disease))
      - Descriptive correlational study
        
        Descriptive study
        To observe/describe a situation
        Measurement of events (behaviour/attitude)
        Obtaining info for the current status of phenomena to describe what "exists" with respect to the variables/conditions
        E.g. opinion poll, prevalence of patients with..., nursing students perceptions
        
        Data analysis
        Descriptive statistics like mean, standard deviation, median, frequency, percentage
        
        Descriptive correlational
        (to find out whether various variables have relationship or not)
        Usually have conceptual framework
        Research questions rather research hypotheses
        
        Pearson's correlation, Spearman's rank correlation, regression, logistic regression
      - Cohort study
        
        Retrospective
        
        Take data from the past to now
        E.g. Nagasaki atomic bomb survivors for risk of preleukemia
        Strength
        Inexpensive
        Time-saving
        Able to study longitudinally with long latent periods
        Weakness
        Risk of confounding variables
        Sometimes unable to determine exact exposure
        Measurement error
        Selection bias
        
        Ambi-directional
        
        Combination of both prospective and retrospective
        E.g.
        Strength
        Captures both short term and long term
        Weakness
        Costly and lengthy
        Risk of confounding variables
        Exposure determination problems
        
        Prospective
        
        Start now, then follow them over the years
        E.g. smoker vs non-smoker grp
        Strength
        Control measure exposure
        Able to measure confounding variables
        Avoid recall bias
        Able to calculate relative risk for the outcome
        Weakness
        Costly and lengthy
        
        Calculate 'incidence rate (IR)’ and then compare
        
        Relative risk = Incidence rate (exposure)/ incidence rate (no exposure)
        
        Beneficial effects RR <0.1 Harmful effects RR > 10
  - - - Inferential statistics: draws conclusions from a sample of data that are subject to random variation like sampling error
      - Statistical Modelling: describe a set of probability distributions by specifying the relationships among one or more random variables and other (non-random) variables; the relationships are usually given as mathematical equations
      - Descriptive statistics: using indexes such as mean or standard deviation
        
        Provide simple summaries about the sample and about the observations
        Form basis of initial description of data as part of a more extensive statistical analysis
        
        CATEGORICAL VARIABLES
        Frequency: count the numbers for each level of a variable
        E.g. Sample of 80 subjects: 45 male (56.3%), 35 female (43.8%)
        
        Percentage
        Cross-tabulation: to show the frequency of one category variable to another categorical variable
        Pictogram: use of graphics to show
        Bar chart/Cluster bar chart/Pie chart/Word cloud
        
        CONTINUOUS VARIABLES
        Mean: Average
        Median: Middle value
        Mode: occurs most frequently
        
        Central tendency
        Symmetric distribution graph
        Median is always central location
        Symmetric graph --> Use MEAN for central location
        Skewed graph --> Use MEDIAN for central location
        
        Variability/dispersion
        Range: difference between maximum and minimum (E.g. 90-160)
        Interquartile range: difference between third quartile and first quartile of data (E.g. Q1-Q3)
        Variance: (data - data average)^2/ (number of data) -1
        Variance = 0 (all data identical)
        Small variance = data close to each (also closer to mean)
        Large variance = data spread out
        
        Standard deviation: square root of variance
        Boxplot; Median: is the line in between the box
        
        Normal distribution graph
        Using mean and variance, we can draw the distribution
        
        MEAN, MEDIAN AND MODE are the same for normal distribution as it is symmetric
        
        mean - standard deviation, mean + standard deviation = 68% etc.
    - - Ordinal
        
        Similar to nominal, BUT there is ranking to numbers.
        No numerical value
        E.g. Pain rating score 0: no pain, 1: mild pain, 2: moderate pain, 3: severe pain
      - Interval
        
        Numerical value meaning
        NO absolute zero present
        E.g. Temperature
      - Nominal
        
        Merely placing levels into separated categories
        No numerical value meaning
        E.g. 1. Male 2. Female
      - Ratio
        
        Numerical value meaning
        ABSOLUTE ZERO present
        E.g. weight
  - - - Probability/Random sampling
        Everyone has equal chance
        
        Simple random sampling
        With sampling frame, k subjects are drawn from a population
        Easy to implement, less bias
        Sampling frame is a must
        
        Systematic sampling
        Randomly selecting one element from the first k-elements in the frame and every k-th element thereafter (with a random start)
        Easier to perform in the field
        Suitable for no good sampling
        May have systematic bias (due to non-uniform distribution of elements)
        
        Stratified random sampling
        Separating the population elements into non-overlapping groups called strata, and then selecting a simple random sample from each stratum
        May produce smaller than simple random sampling
        Can reduce cost per observation
        Can measure parameters according to strata
        Sampling list AND characteristics are necessary
        
        Cluster (random) sampling
        A probability sample in which each sampling unitt is a collection/cluster of elements (e.g. primary schools, nursing homes)
        Less costly
        Good sampling frame is difficult for elements but easy for a collection of elements (e.g. primary school students)
        Systematic difference between clusters
        
        How to draw probability sampling?
        Small number
        2 dice
        Numbered ball/card
        Random number table
        Large subjects
        Use computer softrware
      - Non-probability sampling
        May not have same chance to be selected
        
        Convenience sampling
        Selecting most available people as participants
        Easy to get with low cost
        Atypical sample --> may lead to bias and be difficult to generalize the population
        
        Quota sampling
        Identify population strata and figure out proportion (Male vs female), then draw a convenient sample according to the percentages
        
        Consecutive sampling
        Recruit all subjects from an accessible population over a specific time interval or a specific sample size
        Best possible choice when there is a rolling enrollment
        Better than convenience sampling especially for a long time frame
        May introduce serious bias if the time frame is short or there are specific time slots for subjects
        
        Snowball sampling
        Identify (or several) individuals then ask them to help you recruit people they know (useful in recruiting subjects with specifics criteria)
        Enable to recruit potential subjects who are difficult to identify
        Referrals usually share similar characteristics/features as the first batch of subjects
        
        Purposive sampling
        Select the sample by the researcher based upon a variety of criteria including the specialist knowledge of the research issue
        Inclusion of subjects who are able to provide detailed and relevant data/info to the research qn
        Subjectivity of the researcher's decisiojn making, thus significantly threat to the validity of the research conclusion
    - - Observation
        Direct observation/Observer count frequency of behaviour or event at specific location and time
        
        Time sampling
        Specific selection of the time period during which observations will occur (timeframes can be systematic)
        
        Event sampling
        Select integral behaviours or events to observe; need to have knowledge of the occurence of the event or be a position to wait for event (e.g. violent behaviour)
        
        Important considerations
        
        Focus, defined clearly - broadly defined events
        
        Concealment: try not letting the subjects know
        
        Recording method - By paper and pencil, counter, video etc.
        
        How to record observations?
        
        Recording sheets and checklists
        
        Observation guides
        
        Field notes
      - Self-report
        Data gathered by questioning the sampled subjects, usually called the respondents or participants
        
        Interview
        
        Self-administered
        
        Open ended
        Use open-end questions
        
        Closed-end question
        Pre-specified response alternatives
        
        MCQ
        Rank-order question
        Forced answer question (which statement most closely represent your view)
        Rating question (Extremely satisfied... Extremely dissatisfied)
        Likert scale
        Visual analog scales
        (Scales are useful to quantify the intensity of individual characteristics/feelings but may suffer response set bias:
        social desirability
        Extreme response
        Acquiescence bias
      - Bio-physiologic measures
        Involve the collection of bio-physiological data from subjects by using specialized equipment to determine physical and biological status of the subjects
        
        In vivo measure
        E.g. use thermometer, ECG
        
        In vitro measure
        Measurements perform outside the organism's body; speciments collected and tested outside the body like blood chemistry
      - Records and documents
        
        Provide framework for research project, develop of research questions and validate study findings
        MOH
        Relatively inexpensive
        Provide good sense of background information
        May not have suitable information to the research question
  - - - One sample t-test
        Compare population mean to specific value
        T = specific value - mean/ (standard deviation/square root of n)
        Null hypothesis = x, Alternate hypothesis =/x
      - Independent sample t-test
        Compare 2 population means
        
        both population should follow a normal distribution
        
        both populations have the same variance
        
        Null hypothesis: u1=u2, Alternate hypothesis: u1=/u2
        
        Welch’s tests to adjust so that variance the same
        
        3 more grps?
        
        One-way ANOVA
        
        Hnull = all means equal
        Halternative = not all means equal
        Reject Hnull if p<0.05
        
        Homogeneity is to test for variance
        p <0.05 = Variance not equal (Welch's)
        p > 0.05 = Variance the same (Fisher's)
        
        Post-Hoc Test
        after One-Way ANOVA test, to check pairwise if there is any difference
        Games-Howell (unequal variances) or Turkey (equal variances)
      - Paired sample t-test
        compare 2 means coming from same population
        Null hypothesis: u1-u2 = 0, Alternate Hypothesis: u1-u2=/0
      - ANCOVA
        "..AFTER CONTROLLING for the covariates
        Tests whether there is significant difference between groups after controlling for the variance explained by a covariate
        
        Assumptions
        
        Rs between DV and covariate is linear
        
        Why ANCOVA instead of ANOVA?
        Some variables may have effect on the test (E.g. age in systolic blood pressure of binge drinkers vs non binge drinkers)
        Adjust baseline difference
    - - Mann-Whitney U test
        
        AKA Wilcoxon Rank Sum test
        
        parametric equivalent: independent sample t-test
        Looks at median
        AT LEAST ORDINAL variable
        Hnull: NO difference between 2 grps (median same)
        Halternate: there is difference of variable between 2 grps (median not same)
      - Kruskal-Wallis test
        
        parametric equivalent: One-way annova
        Extension of mann-whitney u test, except 2 or more grps
        Hnull: no difference among the groups
        Halternate: at least one grp is different from the other
        
        Post hoc test (non-parametric) = DSCF (Dwass Steel Critchlow Fligner)
      - Wilcoxon signed rank test
        
        parametric equivalent: one sample t-test
        Looks at median (instead of mean)
      - Wilcoxon signed rank test
        
        parametric equivalent: paired t-test
        Look at median instead of mean
      - Not based on parameterized families of probability distribution
        Include both descriptive and inferential statistics
        Makes no assumption about the probability distributions of the variables being assessed
        
        When to use?
        No clear numerical interpretation (e.g. pain level)
        Used for ordinal variables
        Continuous type data (when the normality of the data is not guarantee or sample size is small)
  - - - Correlation: measures the strength of linear rs between X and Y
        1 refers to perfectly positive linear relation between X and Y
        -1 refers to perfectly negative linear relation between X and Y
        Regression: examine the linear relationship between X and the mean of Y
    - - nonparametric measure of statistical dependence between 2 variables (like pain score and anxiety score)
        Possible values, -1 to 1: perfectly monotonic negative/positive
    - - whether the observations on 2 categorical type variables are independent of each other
        Hnull: 2 variables are independent
        Halternate: 2 variables are not independent
    - - Pearson's correlation coefficient
        
        Measures linear relationship between two continuous type variables
        
        Range from -1 to 1 --> 1: perfect positive linear relation
        H0: p = 0
        H1: p =/ 0
        
        Correlation: measure strength of linear relationship between X and Y
        Regression: examine linear relationship between X and Y
      - Method of least squares
        
        y1 - y1 = residual
        y1: observed value
        y1: predicted value
        Regression line = Least Square line
        
        Before conducting regression analysis, plot a scatter plot to examine relation between 2 variables
      - R: Coefficient of correlation = Pearson correlation (in simple regression)
        R^2: Coefficient of determination: percentage of the variation of Y explained by X, ranging from 0 to 1 (R^2 is monotic increasing with number of variables included)
      - Assumptions of regression
        
        Homoscedasticity: same variance in their errors
        
        Eyeball residual plots
        
        Independence: the response variables are uncorrelated with each other (potential source, dependent variable is time dependent)
        
        Durbin-Watson test
        More than 1 = no autocorrection = independent
        
        Linearity: X and Y are linear related
        
        Scatterplot (Y against X)
        
        Correlation coefficient
        
        Normality (of residuals)
        
        Shapiro Wilk test
        H0: Normally distributed
        H1: Not normally distributed
        If p-value <0.05, statistically significant = reject H0, accept H1 = Not normally distributed
        
        Random sample of population
        
        No measurement error (unrealistic)
    - - Multiple regression
        
        Most widely used statistical method
        More than 1 factor affect dependent variable
        Takes the form of y = a + B1x + B2x + B3x...
        
        Adjusted R2 is used to compare and not only R2, because R2 is monotonic (will increase if variables increase)
      - Model selection
        
        K independent variables can result to 2^k regressions
        
        Backward selection method include all the potential independent variables, and remove insignificant one (largest p-value) one by one until all variables are statistically significant
        
        Use of model builder to compare
        H0: No difference
        H1: With difference
        
        Need to check all assumptions
        No multi-collinearity
        
        Check using Variance Inflation Factor (VIF)
        VIF > 10 --> Multi-collinearity exists
        
        Multi-collinearity – a phenomenon occurs when some independent variables are highly collinear, i.e. linearly related, can cause problems in estimating the regression coefficients
      - Categorical variable in regression
        
        Dummy variable/reference group is made using the smallest number
        E.g. Gender (1-Male, 2-Female)
        Dummy: 0-Male, 1-Female
        
        If 3 categories, one will become reference
  - - - Overarching goal: identify underlying relationship between measured variables
      - Calculate correlation matrix of all k variables to be used in the analysis and calculate the eigenvalue of each k factor
        
        Factor extraction: determine how many factors should be retained
        
        Rotation: rotate the factor matrix to identify the most plausible factor solution
        
        Name the factors and provide interpretation for the factors
        
        Assumption check (check for sphericity)
        KMO greater than 7 --> Good
        Bartlett test: p-value < 0.05 --> statistically significantly different
        
        Scree plot: Eigen value > 1 --> 1 variable
        Also, number of factors may be based on theory/other rationale
  - - - Test-retest reliability: assesses the degree to which test scores are consistent from one test administration to the next
        
        Measured using Pearson's correlation coefficient (correlation between separate administration high --> high test-retest reliability)
        
        However, unable to detect systematic error
        Use of Intra-class correlation (ICC) which ranges from 0 to 1
      - Inter-rater reliability: measures the degree of agreement among raters
        
        Measured using Kappa statistic (or just overall agreement)
      - Internal consistency: reliability assesses the consistency of results across items within a test
        
        High internal consistency = high inter-relatedness of items in questionnaire
        
        Measured using Cronbach alpha: more than 07 is good!
    - - Content validity: estimate of how much a measure represents the construct
        
        Measured using Content Validity Index(CVI): proportion of total items that are judged valid MINIMUM OF 5 EXPERTS
      - Construct validity
        
        Predictive validity: to test a group of subjects for a certain construct and then comparing them with the results obtained at some point in the future
        
        Example: IQ test for university students and the salary 10 years after graduation
        
        Concurrent validity: the scale against benchmark test/scale --> high correlation indicates strong concurrent validity
        
        Example: answer a well-known questionnaire and answer a new scale. If both are highly positively correlated --> concurrent validity
        
        Convergent validity: tests that constructs that are expected to be related, are in fact, related
        
        Example: Select few other scales that are expected to be associated wiwth the target scale and then ask the subjects to fill them all. Examine correlation between them
        
        Discriminant validity: tests that constructs that should have no relationship do, in fact, not have any relationship (also referred to as divergent validity)
        
        Examples: select other scales expected to be NOT associated with target scale and ask the subjects to fill them all. Examine correlation (Discriminant validity should be close to 0 --> good)
      - Face validity: measure of how good scale is at 'face-value' & degree that respondents/personnels judge that the items of an assessment instrument are appropriate to the targeted construct and assessment objectives
        
        Example: If a test is designed to measure whether children are good spellers, and the parents are asked whether the test is a good test, this measures the face validity of the test
  - - - Risk: probability of having an event
        E.g. 2/5 chance of getting cancer --> 40% chance of getting cancer
      - Odds: compare events with non-events
        E.g. horse wins 2 races out of 5 --> odds to winning are 2:3
        
        Learn how to calculate using a table (refer to word doc)
    - - Linear regression takes the form of a formula (y=a +Bx)
        Logistic regression provides binary variable (Death Y=1, Alive Y=0)
        
        Linear regression: Y would increase for Beta unit when X increase 1 unit
        Logistic regression: interpret e^b/odds ratio and its 95% confidence interval and p-value
      - Logistic regression is used to determine which variables affect the probability of a particular outcome or event
    - - Explore associations between one (dichotomous) outcome variables and two or more exposure variables (which may be continuous, ordinal or categorical)
        E.g. how does gender affect the probability of having hypertension after accounting/adjusting for age, income etc. (Accounting for covariates or cofounders = adjustment)