1232334
| Question | Answer |
| Name Steps of the Scientific Method (8 Steps) the Queen Gathered Hippos Consequently After Concluding that they Pee Red wine | 1) Question 2) Gather Literature and Information 3) Form hypothesis 4) Collect and conduct data 5) Analyse results 6) Form Conclusion 7) Publish/Peer Review 8) Replicate |
| What do errors tell us? How do we measure errors? | Errors help assess your prediction and determine how well your data fits your model Calculate Sum of Squares |
| How do we make the SS independent of the data points? What is another term for Mean Squared Differences? What is another term for Mean Squared Errors? | We use the average SS error (MSE) Variances Error Variances |
| What is the "Degrees of Freedom"? | The number of observations that are free to vary Generally... df = # of values - # of parameters |
| Which degrees of freedom is the least likely to be an accurate representation of the population parameter? a) 22 b) 40 c) 2 d) 3 | c) 2 Because it is the smallest number The more data points you have, the better |
| What are outliers and how does it affect our data? | A score that is very different from the rest of the data It affects parameters estimates (description of our population) and the SSm, so it forms biases |
| How do we spot biases? (Violations of Assumptions x 4) ANOVA and regressions are based on the General Linear Model (GLM) = assumptions based on a straight line | Additivity and linearity Normality Homogeneity/homoscedasticity Independence |
| What do we assume with Additivity and Linearity | Relationships between different components in the model are summed up by a linear equation i.e. we assume our relationship will be a straight line |
| What are the three contexts in which we assume normality? What is the rule of thumb when determining a sample size? | Context 1) we test the goodness of fit of a normal model to data - if fit is poor then data is not well modeled Context 2) data that is tested against the null hypothesis - if p < .05, we assume data is not normally distributed Context 3) computes the likelihood that data comes from a normal distribution with given parameter, and then compares that to other variables Sample Size = 30 but it depends on the research |
| What is the z-score? (also called the standard score, and sigma) | The number of standard deviations away from the mean! |
| What does Homoscedasticity/Homogeneity of Variance assume? | The spread of scores of each group are similar; violation will creates bias in estimation of SE |
| What do we assume with regard to Independence? What test do we use to determine this? -you can use this test when your data involves time! | Errors should not be related to each other A Durbin-Watson test (mostly relevant for regression) will determine this for us. |
| What are P-Plots and Q-Q Plots? | Graphical ways of looking at your data |
| How does Skewness and Kurtosis bias your data? | Skewness shifts the data from left to right. A symmetrical distribution has a skewness of 0; a +ve skew has a tail to the right; a -ve skew has a tail to the left Rule of Thumb: if skewness is > 1 or < -1, then skewness and distribution is not symmetrical Kurtosis flattens or sharpens the data. It quantifies the shape; a Gaussian distribution should equal 0; kurtosis that is +ve will sharpen the data; kurtosis that is -ve will flatten the data. |
| What is the Kolmogorov-Smirnov test and when would we use it? | Kolmogorov-Smirnov determines if two sets of data are significantly different "Goodness of fit" Makes no assumption about distribution of data - IT'S FOR NONPARAMETRIC |
| What is a Levene's test? - significance will become more prevalent with larger samples | It assesses the equality of variance (tests the null hypothesis that population variances are equal) If the result is > .05, then we can conclude that differences in sample variance occurred based on random sampling from a population of equal variances |
| If you see a funnel... But if you see a curve... | ...there is a problem with heterogeneity of variance; an assumption is violated ...there is a problem with heterogeneity of variance; both assumptions have been violated |
| Name four techniques to reduce bias: | 1) Trimming the data 2) Transformation of the data 3) Winsorizing 4) Bootstrapping |
| What will happen if Skewness values equals... a) -.099 b) 1.45 c) .02 d) 0 | a) Slight shifted to the right (-ve values shift to the right) b) A relatively large shift to the left (+ve values shift to the left) c) Only very slight shift to the left d) no shift (no bias) |
| What will happen if Kurtosis values equal to... a) -.410 b) .670 c) 0 d) -.980 | a) the distribution curve will appear flatter (-ve values flatten the normal distribution curve) b) the curve will appear pointy (+ve value sharpen the normal distribution curve) c) normal curve (no bias) d) the curve will appear very flat |
| What is homogeneity of variance? | Assumes your range and spread of data are about the same (This implies that is also needs normality and additivity) |
| To reduce bias, how do we trim data? | Trimming data means that outliers are ignored. If possible, try not to trim data when it's not necessary, because you're ultimately deleting data which is'nt ideal. Sometimes SPSS does this by removing the top 5% and bottom 5% of all data. |
| What does Winsorising do? | Transforms data by replacing extreme cases of data - this is done by setting a limitation at each tail ends i.e. 90% Winsorising replaces the bottom 5% of values equal to the minimum value of the 6th percentile, whilst replacing the top 5% values equal to the max value of the 95th percentile |
| What is bootstrapping? | "Resampling the sample data" e.g. in a sample of 100 participants, bootstrapping will take any 50 of those participants and create a random distribution |
| When do we use bootstrap (3)? | When theoretical distribution of a statistic is complicated or unknown When sample size is insufficient When power calculations have to be performed |
| What does transformation do to our data? | Transformation alters data to fit the shape of a normal bell curve |
| When should we use nonparametric testing? | Although generally weaker tests (as they do not use assumptions), nonparametric tests are good when numerical values are only good for ranking. We can then compare certain rankings of different data sets (e.g. average rankings by females vs. average rankings by males) |
| When do we use parametric tests over nonparametric tests? | When the DV is determined as numbers, rather than ranks (the average number is meaningful; nonparametric does not care for averages) When the DV is normally distributed (i.e. assumptions of violations are notable; nonparametric DV may be far from normal distribution) |
| What is Mauchly's test of Sphericity? And what outcome is best for your testing? | Mauchly's test of Sphericity is used for Repeated-Measures ANOVA. It assumes that variances between all possible groups are equal. Thus, if the result is significantly different, this suggests that sphericity has been violated - think homogeneity of variances |
| What is the difference between validity and reliability? | Reliability: being able to reproduce the same results; can this instrument be interpreted consistently across different situations? Validity: coming as close to the true value as possible; does this instrument measure what it sets out to measure? - you can’t have validity without reliability, but you can have reliability without validity |
| What is the difference between experimental and correlational research? | Experimental research manipulates one (or more variables) to see how it affects another Correlational research observes the natural phenomena between two (or more) variables, without interference |
| Two types of variation explain participants’ performance differences? Name these two and explain how they’re different | Systematic Variation: variation of outcomes contributing to a known factor i.e. an experimental manipulation Unsystematic Variation: variation of outcomes due to an unknown factor e.g. time of day, natural abilities of Ps |
| How do you keep non-experimental variation to a minimum in an experiment? | Randomization ensures that if any effect were to occur that is different from another group, this can be attributed to the manipulating variable (the IV) and not to a specific characteristic of that group |
| What does counterbalancing achieve? | Counterbalancing ensures that an outcome is not due to any order effects in the design of the experiment I guess this is also another way of minimizing unsystematic variation |
| Do we always need to transform data? | Not if values can be measured between 0 and 100 i.e. doesn't go any less than 0 or more than 100.... So percentages, counts, latencies, and questionnaire values do not need to be transformed. There will be no outliers. |
| What is the Shapiro-Wilk Test and when would we use it? | Shapiro-Wilk measures null-hypotheses that data is normally distributed, so if p < .05, we reject the null hypothesis, and it's not good Has best power; better for samples < 50 |
| Why do we standardized data? | Standardizing data changes the mean to a 0, so it makes it easier for us to see |
0 comments
Want to create your own Flashcards for free with GoConqr? Learn more.