Help
Options
Didn't know it?
click below
click below
Knew it?
click below
click below
Don't Know
Remaining cards (0)
Know
retry
shuffle
restart
0:00
Embed Code - If you would like this activity on your web page, copy the script below and paste it into your web page.
Normal Size Small Size show me how
Normal Size Small Size show me how
Research Skills B
Quantiative exam revision University level
| Term | Definition |
|---|---|
| Correlation | Relationship between 2 variables which is either positive or negative |
| Correlation part 2 | Closer to 1 the stronger the relationship Does not say something causes something else |
| Linear regression | allows us to predict one variable (DV) from a series of other related variables Builds a model of predictors |
| Regression | Statistical technique that allows someone to predict someone's score on one variable from their score on either one variable or multiple. |
| Criterion vs predictor | Criterion = dependent variable ( what you measure) Predictor= independent variable (what you change) |
| Multiple regression (MR) | Builds a hypothetical model of a relationship between a single criterion variable and predictor variables It is a predictive model that predicts an outcome and produces a regression equation |
| Requirements for multiple regression | 1) Data exploring linear relationships between the predictor and dependent variable |
| Requirements for multiple regression | 2) The DV should be measured on a continuous scale |
| Requirements for multiple regression (MR) | 3) The predictor variable should be measured on a ratio, interval or ordinal scale. ( can also be nominal if dichotomous- splits in two) |
| The different hypotheses for MR | Null= There is no linear relationship between the criterion variable and the predictor variables RH= there is a linear relationship between the criterion variable and at least one predictor variable Research is done to falsify the null hypothesis |
| Assumptions: Sample size | Tabacknick and Fidell (2007) N> 50 +8M where m is the number of predictors The more data the more representative it is. |
| Assumptions :Multicollinearity | The predictor variables should be correlated with the DV but not with each other. A high inter-correlation between predictors is multicollinearity. Singularity =a perfect linear relationship between variables. |
| Assumptions :Multicollinearity | Tolerance values must be above 0.50 Variance Inflated factor (VIF) below 10.00 Correlations between predictors below 0.9 |
| Assumptions :Normality | It is calculated from the residual. Residual is the gap between the actual score and what the model thinks they would score Residual is 0 when the model is perfect |
| Assumptions :Homoscedasticity | Relates to the spread of scores Want them to be evenly distributed DO NOT WANT ALL DATA TO ONE SIDE OR A BOW TIE SHAPE |
| Assumptions: Linearity | The predictor variables should be linearly related to the criterion variable |
| Standard multiple regression | Used to to explain how all predictors together impact the DV |
| Different tables: variables entered/removed | Tells us what variables are included in the model ( relationship between the predictors and DV). |
| Different tables: Model Summary | R= multiple correlation coefficient between predictors R squared= a measure of how of the variability in the outcome is accounted for by the predictors |
| Different tables: Model Summary part 2 | Adjusted R square= gives an idea of how well a model generalises |
| Different tables: ANOVA table | Tells us whether or not the model is a a significant predictor of the outcome variable |
| Different tables: Stanardised coefficients table | Compares the different predictors Look at sig column if it is less than <0.05 then it is significant Can be used for direct comparisons Takes into account what we measure is on different scales |
| Week 1-3 reading part 1 | Regression uses a correlation to predict values of one variable from another Predicted values are found on the regression line on the scatterplot |
| Week 1-3 reading part 2 | Regression is used when you have a specific aim of predicting values on a criterion variable from a predictor variable Multiple regression is used when we have a set of variables which all relate to the criterion variable |
| Week 1-3 reading part 3 | It is a way to partial out the effects of a third variable This is done through a statistical test Can be used to support or challenge a theory between the relationship between confounding variables |
| Regression line | Regression line= line of best fit on a scatterplot which minimizes residuals |
| Multiple regression | Multiple regression = analysis in which the value of one "criterion" variable is estimated using its known correlations with several other predictor variables |
| Linear regression | Linear regression= procedure of predicting values on a criterion variable from a predictor or predictors using correlation |
| Regression coefficient | Regression coefficient= amount by which predictor variable values are multiplied in a regression equation in order to estimate criterion values |
| Stanardised regression coefficient | Stanardised regression coefficient = full name for beta values in multiple regression Beta value= stanardised b weights |
| Predictor | Predictor= variables used in combination with others to predict values of a criterion variable in multiple regression |
| Collinearity | Collinearity= the extent of correlations between predictor variables |
| Multiple correlation coefficient | Multiple correlation coefficient = value of the actual values of the criterion variable used in multiple regression + predicted values. |
| Heteroscedasticity | Heteroscedasticity= degree to which the variance of residuals is not similar across different values of predicted levels of the criterion |
| Week 2 Hierarchical multiple regression | The regression equation allows us to predict the value of the criterion variable (y) from a set of a predictor variables Allows us to predict how Y will change as a result of changes in X. |
| Regression equation | We get the necessary data for the regression equation from the unstanardised B column in the coefficients table |
| Assumptions for multiple regression | Sample size= too few participants and you will be underpowered to find a significant result Multicollinearity= limits the size of r/ makes determining the importance of a predictor difficult |
| Assumptions for multiple regression part 1 | Normality= want peak at 0/ data should be lineally related to the criterion variable Homoscedasticity= residuals at each level should have the same variance/even spread of data |
| Outliers | A piece of data that is substantially different from the rest of the data. Can bias results Tabacknick & Fidell (2001)= values that fall outside of the +/- 3.3 |
| Cook's distance | Measures the influence of deleting a case Should be below 1 |
| Hierarchical regression part 1 | Predictors are entered in a particular order This order is important and is based on previous research They are added in blocks |
| Hierarchical regression part 2 | Each block are assessed on what it adds to the criterion variable Known predictors are added to the first block and newer predictors are added in the other blocks |
| Hierarchical regression part 3 | Requires continuous variables |
| Hierarchical regression part 4 | Stanardised coefficients table= tells us the number of standard deviations that the outcome changes when the predictor changes by one standard deviation MUST CONSIDER WHETHER A HIGH SCORE IS GOOD OR BAD |
| Hierarchical regression Summary | It allows us to estimate the predictive value of each predictor and eliminates confounding variables The regression equation can be useful for making predictions about the value of the criterion values. |
| What can we do with outliers ? | replace value, delete value, non-parametric test, delete participant or transform variable |
| Stepwise multiple regression | A regression method that adds multiple predictors while simultaneously removing that don't improve the r squared value. SPSS only uses the strongest predictors of variance in the outcome variable |
| Stepwise multiple regression part 2 | Variables are added to the regression equation all at once with an attempt to maximize the r squared |
| Stepwise advantages | Saves time = identifies most important predictors Useful for exploratory data analysis=identifies potential important predictors that can be further investigated |
| Stepwise disadvantages | Overfitting= variables that are selected based on statistical significance may generalise poorly to new data ( very specific to the data collected) Biased estimates= produces biased estimates and incorrect conclusions. |
| Stepwise multiple regression use | It can be used to determine what predictors best fit the model and which predictor achieves the highest r squared |
| Summary of regression methods | Standard= force predictors simultaneously. Hierarchical= selecting predictors based on previous work. |
| Summary of regression methods part 2 | Stepwise= bases the order in which predictors are entered into the model on statistical criterion. Best predictor is entered first. |
| ANOVA uses reading | Used to test for differences in lots of experimental designs. Used to analyse data from experiments with more than one IV One way ANOVA is used when there is only one independent variable but more than 2 groups/ conditions |
| ANOVA background | Factorial ANOVA allows for data to be collected from designs with more than one IV One way ANOVA parametric test used to establish whether the means of experimental conditions are different or not Factors are also known as the IV |
| ANOVA | Can compare participants performance across multiple groups/ conditions Tells us whether scores vary across conditions Does not tell us whether conditions significantly vary across conditions |
| F ratio | Variance due to manipulation of IV divided by error variance |
| Level of factors | A single IV that is manipulated to create multiple conditions |
| Between subjects factors | Factors that vary between participants and participants only experience one level of a factor |
| Within subjects factors | Factors whose levels vary within a participant who do all conditions. |
| Main effect | Effect that has a single independent (IV) has on a dependent variable (DV) |
| Intro to ANOVA (Analysis Of Variance) | Can cope with lots of independent variables It is like an extension of a t test |
| What is ANOVA and how is it used ? | Parametric test used to test variability Used when have MORE THAN two groups Conditions are also known as levels/ the independent variable are known as factor |
| Advantage of ANOVA | Can see the effect of multiple factors on DV at the same time (look at combined effect of the variable Efficient for more than 2 groups |
| Assumptions of ANOVA | The DV consists of data measured at interval or ratio level |
| Assumptions of ANOVA part 2 | The data for the DV is normally distributed |
| Assumptions of ANOVA part 3 | There is homogeneity of variance |
| Assumptions of ANOVA part4 | For independent group designs, independent random samples must have been taken from each population |
| Levels of measurement | Nominal Ordinal Interval Ratio |
| Nominal data | Numbers given to distinguish between categories with no particular order to rank importance e.g ethnicity, gender etc |
| Ordinal data | Numbers are given to distinguish between categories but order is importance. e.g 1st, 2nd etc |
| Interval data | USED IN ANOVA Put scores in order with equal distances between intervals e.g temperature There is no true zero ( 0 degrees does not mean that there is a absence of temperature, the difference between 10 and 20 is the same as 0 and 10). |
| Ratio data | USED IN ANOVA Like interval data but with a true zero e.g years of work, number of children in the household The difference between 2 & 4 is the same as 0 & 2 but 0 is the absence of children |
| Normally distributed data | The data is drawn from the population which is normally distributed DV is normally distributed Bigger sample sizes are more likely to be normally distributed |
| Homogeneity of variance | Variance within a sample mean does not have to be the same but should be. |
| Independent random samples | They are taken from each population |
| How does ANOVA work ? | It analyses different sources which variations in scores arise Looks at variability between conditions (between- groups variance) and within conditions (within- group variance) |
| Between - groups variance | The variance BETWEEN group means This can be caused by three things : individual differences, treatment effects, random effects |
| Within- groups variance | The variation of people WITHIN the same group/ the difference in means within the same group We want this variance to be small Also known as error variance |
| Error variance | Variability within the group which is not produced by the IV It only includes: individual differences and random effects |
| Individual differences | People naturally vary We want to avoid high amount of individual differences as it can make it seem that the IV is having an effect when it isn't This is why random allocation is used. |
| Random effects | Errors in measurement that arise from a range of sources : varying external conditions, state of the participant etc |
| Treatment effects | The effect of the IV(s) / what we are trying to measured Variance due to having different groups of people This is anticipated. |
| The logic of ANOVA PART 1 | Subjects in different groups should have different scores because they have been treated differently ( between- group variance) |
| The logic of ANOVA PART 2 | Subjects within the same group should have a similar score ( within the same group) |
| Null hypothesis for ANOVA | The populations from which the sample have been drawn have equal means. There will be no difference in test scores..... |
| Research hypothesis | The populations from which the samples have been drawn do not have equal means. There will be difference in test scores... |
| F ratio | ANOVA calculates the ratio of variance due to our manipulation of the IV and the error variance This should be below 1 |
| when is the f ratio significant | If above 1 we decide if value is large enough to be statistically significant but p has to be below 0.05 for it to be significant If error variance is large then f will be above 1 |
| ANOVA VS regression | MR is used to predict a continuous outcome on the basis of one or more CONTINUOUS predictor variables ANOVA is t is used to predict a continuous outcome on the basis of one or more CATERGORICAL predictor variables |
| Mixed ANOVA | It is used when a study includes one or more within- subjects factors and one or more between subjects factors. |
| Ways to describe an ANOVA | How many FACTORS are in the design How many LEVELS are there in each factor ( e.g. time of day has 3) Within or between subjects |
| Levene's test of equality of error variances | Tests the homogeneity of variance assumption |
| Mean square | Each sum of square is converted into an estimate of variance by dividing it by its degree of freedom |
| Mean square part 2 | between groups sum of squares divided by between groups degrees of freedom |
| Mean square part 3 | within groups sum of squares divided by within groups degrees of freedom |
| Main effect | It is reported as F, df, df error, F value, p value |
| Effect size / eta squared | Tells us how much variation in recall scores can be accounted for by the the conditions For a one way ANOVA partial eta squared and eta squared are calculated the same. 0.01 smal, 0.59 medium, 0.138 large |
| calculation for eta squared | sum of squares for between groups condition divided by total sum of squares |
| Planned ( a priori) comparison | Conducted when the researcher has hypothesised which means will differ from each other in advance The overall main effect does not need to be significant |
| Planned and unplanned comparisons | They are used when there is a large difference and we do not know where this difference is |
| Unplanned ( post hoc) comparison | Differences in means explored after data has been collected Don't use if overall main effect is not significant (USE IF SIGNIFICANT) |
| Contrast tests | Tells us which means have been compared |
| Cohen's d effect sizes used in comparisons | 0.2 small 0.5 medium > 0.8 large |
| When should we use these tests | a factor has more than 2 levels The main effect is statistically significant |
| Unplanned ( post hoc) comparison part 2 | SPSS adjusts for multiple comparisons being made and corrects to reduce a type 1 error |
| Type 1 error | When the null hypothesis is rejected when it should have been accepted/ is true. Bonferroni's adjustment is used to reduce this in this circumstance |
| One way ANOVA | means that there is only 1 factor Should always show what a high score would suggest |
| Advantages of repeated measures design= increased stats power | Increased statistical power/ removes the effects of individual differences = using same group of participants so there is no difference between them |
| Advantages of repeated measures = less time & money | less participants needed --> less time and money needed |
| Disadvantages of repeated measures design = Pratice effects | Practice effects= people get better the more they do a task Fatigue= get tired/ bored of doing the same task |
| Disadvantages of RM design = contrast effects | Contrast effects= having two tasks together may impact each other Demand characteristics = people try to guess the aim of the study and act accordingly |
| Solutions to order effects = Randomisation | Randomization of order testing= people are randomly assigned. May end up with more in one group compared to another |
| Solutions to order effects= Counterbalancing | Counterbalancing= half do A then B, other half do B then A |
| Repeated measures vs independent groups | RM is superior as there is no variance between the groups |
| Sphericity | Replaces homogeneity of variance in independent groups ANOVA. The variances of the difference between all possible pairs of conditions |
| Factors and descriptives | Within subjects factors = displays the levels of the factor Descriptive stats table |
| Mauchley' s test of sphericity = | Assumption of sphericity shown in this table Want result to be above. 0.05 Within subjects design with more than 2 levels |
| Test for within subjects effects | F( first df,second df) = f value, p value, effect size |
| Effect size | How confident we are that we can reject the null hypothesis & generalise to the wider population Need to calculate for a 2 way ANOVA It is stanardised / allows for comparison between groups |
| Effect size in more detail | The size of the difference between groups Small effect sizes = are not visible to the naked eye |
| Eta squared IM Design | Sum of squares for factor divided by sum of squares for factor + sum of squares for error 0.138 is a large effect/ difference between meane |
| Mean affect | Looks at the effect of each of the factors on their own Can be used to compare factors |
| Interaction | The combined effect of the factors A significant interaction = when one factor differs |
| Levene's test of equality of error variances | Replaces assumption of homogeneous of variance in independent group designs |
| How can you see an interaction on a graph | lines parallel = no significant interaction lines not parallel= significant interaction crossover= means met in different situations |
| When is planned and unplanned comparisons used | when a factor has three or more levels they are used Planned= when we have hypothesised the means are different Unplanned= when we have not hypothesised the means being different |
| Main effect is significant | if have specified hypothesis= run planned comparisons Have not specified hypothesis= run unplanned comparison |
| Main effect is insignificant | Specified hypothesis= run planned comparison Have not specificed hypothesis= take no further action |
| One way vs two way ANOVA | One-way = 1 IV 7 DV Two-way = 2 IV & DV, looking at individual & combined effects on DV |
| Between subjects (BS)/ independent measures RECAP | BS= factors that vary between participants/ pps only experience 1 level WS= factors vary within participant |
| Mixed ANOVA | one or more factor with same pps= within one or more factor with different pps= between |
| Mixed ANOVA tests | Main effect of each of the factors An interactive effect of the factors |
| Sphericity in Mixed ANOVA | the variances of the differences between all possible pairs of conditions ( levels of factors ) are equal Within subjects- Mauchly's test |
| Homogeneity of variance | Samples are compared to populations with a similar variance Between subjects= Levene's test |
| Box's test of equality of covariance | For each level in BS factors , the pattern of intercorrelations should be the same |
| Parametric test | based on the estimation of certain parameters related to a certain population Parameter= characteristic of a population |
| When to use non- parametric test | data is non-normal Outliers ( that you do not wish to remove) DV isn't ratio or interval Test- specific assumptions are violated Small sample size Unequal sample sizes if using groups |
| Mixed ANOVA variance assumption | Mauchly= within subjects Box test when both is used |
| Why do we not use parametric tests more frequently | They are more powerful They are more likely to find differences/ relationships than parametric tests |
| Kruskal Wallis | Non- parametric equivalent to one way between subjects ANOVA Tests for significant differences in two or more groups of an IV/factor on a continuous or ordinal DV |
| When is Kruskal Wallis used | Non- normal distribution of data Interval, ratio or ordinal data Small sample size or unequal groups Outliers |
| Follow up test to Kruskal Wallis | Mann Whitney is used as a non-parametric alternative to an independent t- test Bonferroni adjustment (0.05) is used |
| Problems with Kruskal Wallis | It assumes that The samples are randomly sampled from the population Each value is obtained independently |
| Friedman test | Non- parametric equivalent to one way within subjects ANOVA |
| When is the Friedman test used | Non- normal distribution of data Interval, ratio or ordinal data Small sample size Outliers |
| Problems with Friedman test | It assumes that the sample was randomly sampled from the population |
| Friedman test | Tests for significant differences in two or more groups of an IV / factor on a continuous or ordinal DV |
| Follow up test Friedman | A sign test is used Bonferroni's adjustment (0.05) |
| p value | How likely it would be to get the pattern of data we have found if the null hypothesis were true if p is less than 0.05 = not enough evidence to reject null hypothesis p more than 0.05= have sufficient evidence to reject null hypothesis |
| p value explained | if p is less than chosen significance level then null is rejected for alternative hypothesis |
| Type 2 error | when the null hypothesis is accepted when it should have been rejected |
| Alpha & beta | Alpha = probability of committing a type 1 error set at 0.05 (5%) Beta= probability of committing a type 2 error set at 0.2 (20%) |
| Bonferroni adjustment | reducing the probability of making a type 2 error Divides acceptable probability level by the number of comparisons we wish to make. |
| Power | the ability of a test to detect an effect The probability will find an effect assuming that it exists |
| Cohen (1992) power | proposed a 0.2 probability of failing to detect a significant effect Power= 1- beta/ power= 1-0.2 |
| Power scale | 0= no power at all 0.1-0.3= low 0.4-0.6= moderate 0.7 and above= high |
| The importance of power | can assist in the design/methodology of a study Idea of how many people you will recruit can help with funding determines feasibility of study Determines whether or not you are likely to get statistically significant results |
| Three main factors influencing power | size of the effect you expect to find Alpha level Sample size |
| Other factors influencing power | Type of statistical test Design One or two tailed test |
| When is power calculated | prospective /a priori= calculate sample size to achieve wanted level of power post-hoc= calculate power after research has been done |
| Prospective/a priori calculation | Significance level we hope to use (0.05) Power we are hoping to achieve ( around 0.8) The effect size we are expecting ( based on previous research) |
| How to calculate power | Analytical formula Published tables Software |
| When might a post- hoc power analysis be useful | pilot studies to help explain unexpected findings restricted sample sizes Can be useful in directing future research |
Created by:
Adridri1234
Popular Psychology sets