PS2010

PS2010

SPSS
1. stats tests
  1. Chi-squared
    1. analyse nominal/frequency/categorical data
      1. nominal data: describes the group a P belongs to
        NON-PARAMETRIC
  2. Differences between conditions: COMPARISIONS
    1. ANOVA, t test
      1. t tests
        ERRORS
        Type 1: THE WORST!!!!!
        Finding effect in sample, accepting H, but NO effect in real life
        Incorrectly finding signif. effect
        Cannot avoid unless testing ENTIRE population! 5% error allowed ALPHA LEVEL (p <.050)
        Type 2
        Found no effect in sample, accept null H, but there ARE effects in real world
        PARAMETRIC ANALYSIS
        Homogeneity of variance
        LEVENE'S TEST
        Not signif.: (p > .050)
        ASSUMPTION HAS BEEN MET
        "equal variances assumed" row
        p GREATER THAN .050
        Signif.: p < .050
        ASSUMPTION HAS BEEN VIOLATED
        "Equal variances not assumed" row
        p LESS THAN .050
        Independent t test
        SPSS output
        "Independent samples test"
        LEVENE'S
        t (df) = t statistic, p value
        descriptives
        Population (N) individual group (n)
        include M and SD
        Confidence intervals (CI)
        CI of a mean: uses N and SD -> lower and upper score
        Can be 95% certain the upper and lower scores will reflect any sample taken from data set - even if exp is replicated
        WRITE UP
        comparing groups
        "Group stats"
        Null hypothesis - A = B, no direction stated
        After analysis - either accept (if p is NS) or reject null hypothesis (p is Sigif.)
        If SPSS output - p = .000 -> p < .001
        Null hypothesis - A = B, non directional; easiest H to make if there is no previos research
        After data analysis - either accept or reject Null H
        Accept: there is a relationship
        Reject: there is no relationship
        Repeated t test
        SPSS output
        "Paired samples test"
        t (df) = t statistic, p value
        no homogeneity of variance or Levene's test
        violates assumptions of independence of obs.; random variance (WITHIN groups) is reduced
        SPHERICITY ASSUMPTION
        W = W, x2 (df) = x2, p = p value
        One sample t test
        compares data from a single group to a "reference value"
        eg. population
        SPSS output
        "one sample t test" -> test value (reference)
        interpreting direction: is sample M signif higher/lower than test value?
      2. ANOVA: Independent measures
        Analysis of Variance; compares many groups
        3 MAIN ASSUMPTIONS
        normal distribution of DV
        homogeneity of variances
        LEVENE'S
        independence of observations
        SPSS OUTPUT
        "Between subjects factors" + "descriptives"
        "Levene's test of equality of error variances"
        F (df1, df2) = F ratio, p = p value
        if Levene's is NS (p > .050)
        GOOD - Assumption has been met
        Bonferroni
        (few pairwise comparisons)
        adjusts alpha level: .050 / no. of conditions
        Tukey
        (many pairwise comparisons)
        Levene's = signif (p < .050)
        BAD - Assumption has been violated
        Games-Howell
        "Tests of bet. subjects effects"
        F (model df, error df) = F ratio, p = p value
        eg. if IV = time, DV = no. words recalled -> main effect of time is reflected through this write up
        write up
        explain analysis used
        has asumption of homogeneity been met?
        report and interpret dirction of main effect
        "Source" error = random variance
        "mean square" time: explained V; error: random V
        big F ratio = likely signif !!!
        Preferred to multiple t tests because...
        eg. w/ two t tests = 5% error x 2 = 10% error
        increased chance of making a Type 1 error w/ multiple t tests
        Familywise error: making one or more type 1 errors
        F ratio: ratio of explained (experimental) to random variance
        Exp / random
      3. ANOVA: Repeated measures
        directional H - planned contrasts
        non-directional H - post hocs are limited for rep. anova
        use bonferroni
        SPSS
        (don't need to look at "multivariate tests") !
        "within sub. factors" - coding variables for contrast interpretation
        descriptives
        Mauchly's test of sphericity
        if NS: assumption of sphericty has not been violated
        "sphericity assumed" rows - main effect stats
        is S: sphericity been violated
        GREENHOUSE GEISER (alters df)
        W = Mauchly's W, X2 (df) = Chi squared value, p = sig.
        Repeated measures ANCOVA
        EMM's rather than descriptives
        Mauchly's
        Confound: F (covariate df, error df) = F ratio, p = sig.
        IV * confound interaction row can be ignored in SPSS output (usually NS)
        "tests of within subjects contrasts" - ignore interaction and error rows
        G-G if Mauchly's is S
        Two-way repeated ANOVA
        Two-way Mixed ANOVA
        1 (or more) IV that has independent p's+ 1 (or more) IV that has repeated p's
        eg. examining whether the importance of looks (IV1) differs for males and females (IV2). DV: likeliness of going on 2nd date
        SPHERICITY: repeated IV
        LEVENE'S: independent IV
        interpreting main effect + interaction - MEANS and plots
        break down interaction: split file for either looks / gender; repeated ANOVA
        Three-way mixed ANOVA
        3 IV's - independent and repeated measures
        if there are only 2 levels of each IV - ASSUME SPHERICITY
        EMM'S for interpretation
        IV1 main effect
        IV2 main effect
        IV3 main effect
        Interaction effects
        break down w/ split file
        effects of indep. and repeated V's are presented in diff parts of output !
        2 IV's: same p's in different conditions of each IV
        eg. examining whether looks (IV1) and personality (IV2) have an effect on attractiveness (DV)
        SPHERICITY
        W = Mauchly's W, X2 (df) = Chi squared value, p = sig.
      4. ANCOVA
        takes into account a covariate (variance caused by confound)
        when analysisng a covariate - could explain some of the random variance
        SPSS
        EMM's
        adjusted using covariate
        interpreting output requires EMM''s rather than unadjusted descriptives
        "tests of between subjects effects"
        will display 2 results above "error": first will be the covariate, second will be the IV
        was homogeneity of variance met? was the covariate S? was the main effect S?
        covariate must be continuous and binary
        Stage 1
        does the covariate explain a signif amount of variability in the DV
        Stage 2
        after controlling for the covariate, is there more exp V than random V?
        more random V than exp V: ANOVA NS
        covariate explains small amount of random V: covariate NS
        thus, ANCOVA will be NS
        more random V than exp V: ANOVA NS
        covariate explains a lot of the random V: covariate S
        more exp V than random V: ANCOVA is S
      5. Factorial independent ANOVA
        two-way: 2 variables / factors
        different p's in each cond.
        eg. sex and alcohol consumption (IVs) on aggressiveness (DV)
        examines main effects (effect of each factor on its own) AND interactions between the factors
        break down MAIN EFFECTS if 3+ cond.'s
        line graph /plots reveal significance if:
        lines are not parallel (going in diff directions) - but S depends on angle!
        lines are crossing
        2 conditions only need to report MEANS "estimates"
        breaking down interaction effect
        separate independent t tests for each level of IV1 to compare to IV2
        "independent samples t test"
        factorial ANOVA vs one-way ANOVA
        ADVANTAGES
        analysing interaction effects
        adding variables reduces error term - accounting for random variance
        SPSS output
        LEVENE'S
        "tests of between subjects effects"
        "source" - IV1, IV2, IV1 * IV2
        "multiple comparisons" - post hoc output
  3. Relationships between variables
    1. Correlation, regression
    2. Complex correlations
      1. Partial correlations
      2. PEARSON'S r
        r values range from perfect negative (-1) - perfect positive (+1)
        line of best fit - represents DIRECTION of relationship
        residuals: diff betw. raw data point and line of best fit
        smaller residuals - more accurate model - line of best fit reduces random variability
      3. OUTPUT
        "CORRELATIONS": R VALUE, P VALUE, N
        r = .660 POSITIVE relationship
        r = -1.03 NEGATIVE relationship
      4. correlation does not imply causation
    3. Multiple regression
      1. Complex regression models
      2. assumptions
        4
        multicollinearity
        distribution of residuals
        homeoscedacity
        outlier effects
      3. Categorical variables in regression
      4. beyond simple correlations: analysing 2 + CONTINUOUS variables
        PREDICTIVE MODELS
        outcome variable
        predictor variable
        "IVs"
        "DV"
      5. OUTPUT
        PEARSON'S r
        R2 and adjusted R2
        explained variance in outcome V by predictor V
    4. Factor and reliability analysis
  4. Advanced stats
    1. effect size
    2. power analysis
2. BASICS
  1. Data view: enter data
    1. Variable view: define variable properties
      1. Decide if variable is a continuous score or categorical definition
        Continuous data (always changing/ not fixed; IQ, age) : just enter raw data
        Measure: SCALE
        Categorical data (fixed; sex): needs to be coded
        Measure: NOMINAL
        Before analysis: list, name all variables + demographics
      2. Row represents one variable
      3. For catergorical, define VALUES and MEASURE
    2. Row (across) - participant data
    3. Column (down) - variable
3. Planned contrasts
  1. based off directional H
    1. one-tailed
  2. DEVIATION
    1. Contrast 1
      1. 2 vs 1,2,3,4 etc
    2. Contrast 2
      1. 3 vs 1,2,3,4
    3. Contrast 3
      1. 4 vs 1,2,3,4
    4. compares effect of each cond. (except 1st) with overall effect
  3. SIMPLE
    1. compares effect of each cond. to 1st (reference)
    2. Contrast 1
      1. Contrast 2
        Contrast 3
        1 vs 4
        1 vs 3
      2. 1 vs 2
  4. DIFFERENCE
    1. compares effect of each cond. to overall effect of previous cond's
    2. Contrast 1
      1. Contrast 2
        Contrast 3
        2 vs 1
        3 vs 2,1
      2. 4 vs 3,2,1
    3. opposite to HELMERT
  5. HELMERT
    1. compares effect of each cond. to all following cond's
    2. Contrast 1
      1. Contrast 2
        Contrast 3
        3 vs 4
        2 vs 3,4
      2. 1 vs 2,3,4
    3. opposite to DIFFERENCE
  6. REPEATED
    1. compares effect of each cond. to the next cond. only (not all following cond's)
    2. Contrast 1
      1. Contrast 2
        Contrast 3
        3 vs 4
        2 vs 3
      2. 1 vs 2
  7. 6
  8. no. of conditions - 1 = no. of contrasts
  9. POLYNOMIAL
    1. looks at patterns in data
      1. trend analysis only appropriate for continuous IV's !
    2. Linear trend
      1. Quadratic trend
        Cubic trend
        2 changes in direction
        1 change in direction
      2. straight line
4. Post-Hoc
  1. based off non-directional H
    1. two-tailed
Descriptive stats
1. Measures of central tendency
  1. Mode, median, mean
2. Dispersion
  1. Range, variance, SD
    1. Experimental variance: variability between conditions
      1. experimental manipulation
      2. GOOD: likely significaant
      3. large F ratio - significant
        explained by more exp variance compared to random variance
    2. Random variance: variability within conditions
      1. measurement/ human error
      2. Indiv. diffs.
      3. unaccounted for/ unmeasured varaibles
      4. BAD: not likely to be significant
Inferential stats: tell us if data is signif. or not
1. PARAMETRIC
  1. 4 assumptions
    1. Interval / ratio data
      1. interval: - values possible
      2. ratio: - values not possible
    2. normal data distribution
    3. Independence of observations
      1. responses from P's (observations) should not be influenced by each other
    4. Homogeneity of variance
      1. "SAME" pattern of variance in all groups
      2. Positively skewed: tail in + direction (right)
        Negatively skewed: tail in - (left)
        HIGH scores over-represented
        LOW scores over-represented
      3. Ideal because we need to be confident of the mean differences
2. NON-PARAMETRIC STATS
  1. No normal distribution needed!
    1. No homogeneity of variance needed
      1. RANKING DATA
        SPSS analysis is based around RANKS rather than actual data
        report MEDIANS! ("statistics")
  2. Independent
    1. 2 conditions
      1. Mann-Whitney / Wilcoxon Rank-Sum
        Independent t test
        U = Mann-whitney U, z = Z, p = Asymp. sig.
    2. 3 + conditions
      1. Kruskal-Wallis
        Independent ANOVA
        H (df) = Chi-squared value, p = Asymp. sig.
  3. Repeated
    1. 2 conditions
      1. Wilcoxon-Signed-Rank
        Repeated t test
        T = (smallest value under "Ranks" table - "mean rank" column), p = asymp. sig.
    2. 3 + conditons
      1. Friedman's
        Repeated ANOVA
        X2 (df) = chi-squared value, p= asymp. sig.
  4. Testing assumptions
    1. interval/ ratio data
    2. normal distribution
      1. histogram
        skewness
      2. normality tests
        SPSS output
        "tests of normality" - under this will be the name of test to report
        D (df) = statistic, p = sig.
        if S (p < .050) = BAD (significantly different from a normal dist.
        NS = GOOD
      3. Kolmogorov-Smirnov: D(df) = statistic, p = sig.
    3. homogeneity of variance
      1. LEVENE'S (F ratio)
Questionnaire design
1. reliability - consistency
  1. internal consistency of each item (esp. if in subscale): consistent scores
2. validity - is the measure measuring what it claims to?
3. be specific (eg. rather than "regualrly", use "weekly, daily" etc
  1. no double negatives / double barrelled Q's (eg. no two issues in one Q)
    1. give option of not responding to any sensitive Q's
    2. Consistent response options! (eg. strongly agree, agree...)
4. Open Q's
  1. advantages
    1. unrestricted response
    2. detail and additional info- qualitative data
  2. disadvantages
    1. difficult to analyse + summarise responses
    2. time consuming for respondent -> participant effects ?
  3. Thematic analysis
    1. identify main themes that emerge from data
      1. validity: do the themes reflect what the p's said/ meant
      2. inter-rater reliability in analysing data / extracting themes
        reflexivity: awareness that researcher can never be unbiased
    2. Interpretative phenomenological analysis (IPA)
      1. most appropriate for finding answers about the experiences of certain groups
        sampling methods
        purposive sampling: homogenous p's
        idiographic approach: individual cases
        double hermeneutic: p makes sense of experience; researcher understands + interprets
      2. phenomenology: the phemomena that we see in the world around us
        what the researcher brings to the text is important for analysis; engages with p's account of phen. rather than phen. itself
        analysisng p's account of depression rather than depression as a condition
      3. GOAL : generate list of master themes (inclu. p's shared experience and essence of phen.)
    3. most appropriate for finding themes in population
  4. Content anaysis
    1. derives semantic themes from TA and looks at their occurrence
5. Closed Q's
  1. advantages
    1. quick and easy to analsye / code
  2. disadvantages
    1. fixed choice of responses
    2. inter-rater reliabilty and bias in coding
  3. closed responses
    1. categorical (male / female)
    2. likert scale
    3. ranking items in order
    4. possibility of acquiescence bias (always agreeing w items)
Interviews and focus groups
1. identify key themes and terms
2. Unstructured
  1. no set questions
3. Semi-structured
  1. Q's set as a GUIDE
4. Focus group - group interview; interaction between p's is a source of data
  1. similar structure to semi-structured interviews
    1. less artificial than one on one interview
    2. less appropriate for sensitive topics

Nächster

Beschreibung

Zusammenfassung der Ressource

ähnlicher Inhalt

	Erstellt von Jada Khan vor mehr als 7 Jahre