Research Methods Exam 3

t-test reporting

-"An independent-samples t test showed that the mean change in MEP between the groups was not significantly different, t(10) = 0.654, p = .528." -(10) --> Degrees of freedom (N-1 etc.,)

paired t-tests

-"related samples t-test" -to compare the means of a continuous variable in 2 non-independent samples (ie, measurements on the same ppl before and after tx) -any pre-post type of testing (ex. surgery, therapy, session) -or if you are comparing the same person but as session1 vs session 10 -is diet X effective in lowering cholesterol lvls in a sample of 12 ppl? >measurement is on the same ppl but before and after tx -do patients who receive LSVT have higher speech intensity after tx than they did before tx? -essentially the same person taking the test twice

set an acceptable lvl of risk (step 2)

-4 possible outcomes when testing a hypothesis: >H0 is accepted when it is true (correct decision) >H0 is rejected when it is false (correct decision) >H0 is rejected when actually it is true (type 1 error/alpha): what lvl of error are we willing to accept? >H0 is accepted when actually it is false (type 2 error/beta): failing to find an affect that does exist in reality (you can't find it) -Probability of Type I error is designated as alpha. -Probability of Type II error is designated as beta. -Value of alpha is specified before data collection. Beta can be determined only when population parameters are known. So beta is not used in our field as we typically include populations with unknown parameters. -Alpha is conservatively set at .05 or less. When alpha is equal to .05 it means that the researcher is willing to risk a Type I error 5 out of 100 times. When alpha is equal to .01, the researcher is willing to risk a Type I error only 1 out of 100 times. -Type I errors are considered more serious than Type II errors. Risk of rejecting a true null hypothesis (Type I error) is potentially more damaging than making a Type II error.

ANOVA assumptions

-Data are continuously measured (interval or ratio level). -Random selection of participants from the population. -Data are sampled fr populations w normal distribution as estimated by sampling characteristics. However, ANOVA is robust to violations of normality when sample size is larger (n> 30). -Data are sampled from populations with equal variances as estimated by sample variances. ANOVA is especially vulnerable to violations of equal variances when sample sizes are unequal. Levene's test for multiple samples is used to test for equal variances.

guidelines for planning research

-Hypotheses should be two-tailed unless there is compelling evidence for a possible result direction. -Choice of bidirectional or nondirectional hypotheses should be decided before data collection. -Researchers should provide rationale when using one-tailed hypotheses.

t-tests interpretation

-If p < .05, means are significantly different from each other -If p > 0.05, means are not significantly different from each other

Model for clinical outcome research

-Phase 1 research is explorative, based on a tentative tx protocol. phase 1 observations are designed to detect the presence (or absence) of a tx effect as well as any negative consequences. participants may not exactly represent the target population, and external controls may be lacking, thus phase 1 studies aren't rigorous experiments but necessary to establish ground rules -Phase II goals differ from those of Phase I in that they aim to finalize operational definitions, define the exact population of interest, refine methodology, and explore the tx effect's degree and permanency -Phase III clinical-outcome research aims to test the critical hypothesis and answer the research q regarding tx efficacy. Includes large, representative samples of subjects and includes external controls -Phase IV aims to bridge the divide between research and practice. it's particularly important for researchers to collaborate with clinical researchers to implement Phase IV clinical-outcome research. the focus of research in this phase may shift to specific subpopulations or could extend the tx protocol to different populations -Phase V clinical-outcome research shifts to other tx effectiveness issues such as cost-benefit, consumer satisfaction, and quality of life issues. Phase V studies typically involve large-group and single-case designs but usually don't include a comparison group. Systematic reviews are useful for combining the numerous results from Phase IV and V studies.

Standardized effect size interpretation

-The sizes of the effect and standardized effect are correlated to the amount of clinical significance they have (from ch 10 HW) -Cohen (1988) suggested following conventions for interpreting _____ ____ ____: 0.80 = large 0.50 = moderate 0.20 = small <0.20 = trivial

skewed distributions

-When the spread of data is not symmetrical meaning the data clusters to one end. -The mode is located at the highest point, then the median and finally the mean. ex. -number of lottery won -work bonuses -number of suicides

regression

-___ analysis is used to predict the value of one variable (the dependent variable) on the basis of other variables (the independent variables)

normal distribution

-a theoretical distribution which provides a model for evaluating the distributions of real-life variables. -useful for determining the probability of certain outcomes -"ideal" data follows a "normal" bell shaped curve -aka normal, bell, and gaussian

two-tailed hypothesis

-a type of Ha -it doesn't predict the direction of the results -for this hypothesis, the rejection region is equally between the two ends of the distribution. If the expected result is a positive difference, the rejection region is located in the right tail. If the expected result is a negative difference, the rejection region is located in the left tail. ex: a researcher might hypothesize that treatment A is different from treatment B but not predict which treatment is better.

one-tailed hypothesis

-a type of Ha -predicts the direction of results -for this hypothesis, the area of the sampling distribution can be divided into two regions: left/right of the shaded region is the rejection area and the shaded area is the acceptance area. -this is more powerful than the 2-tailed test bc the rejection area is larger for the ___ test ex: a researcher might hypothesize that treatment A is better than treatment B

t-test

-basic level -tells you if there is a statistically significant difference between the mean score (or value) of two different groups of people (e.g. males vs. females; PD vs. controls) or the same group of people before and after tx/intervention. -in essence, this test gives a measure of the difference btwn the sample means in relation to the overall spread -types of tests: independent samples _____ and paired ____ -there may be an overlap in the distributions, where it is unclear whether the samples come from the same population ex. lung capacity of 10 yr olds and 18 yr olds will show 2 graphs that do not overlap at all ex. lung capacities for males and females will show an overlap

ANOVA

-cannot use t-test if you have more than 2 tests (bc it's only for 2 samples) -use analysis of variance (____) to examine multiple comparisons at one time -types of ___ analysis: >one-way ____: one grouping factor (ex. a dx category) >two-way ____: two grouping factor (ex. dx and gender) >three-way ____: three grouping factors -____ w multiple factors is called multifactor analysis of variance

determining critical value (step 4)

-critical value: a cutoff point is used in statistical hypothesis testing which can be used to separate sample results leading to rejection of H0 fr sample results leading to acceptance of H0 -critical value depends on: alpha lvl and alternative hypothesis -2 types of Ha: one-tailed (directional) hypothesis and two-tailed (non-directional) hypothesis -can help us to know when to reject or accept H0

inferential statistics

-data are used to make conclusions about populations (estimating some characteristic of the population based on what we know about the sample) -tells you if data is significant or not -includes making inferences, testing hypotheses, determining relationships, and making predictions -T-tests -ANOVA -chi-square -correlation and regression -inferences abt a phenomena: proving or disproving theories, associated btwn phenomena, if sample relates to the larger population, eg diet and health -the validity of the inference depends on the quality of the sample. if the sample is a good representation of the population, the inference is likely to be good as well

ANOVA computation

-essentially, _____ involves dividing the variance in the results into: >btwn groups variance >w/in groups variance F= (measure of btwn groups variance)/(measure of w/in groups variance) -the greater F, the more significant the results (values of F in standard tables) -Nonparametric alternative to one-way analysis of variance: Kruskal-Wallis test -Nonparametric alternative to two-way ANOVA is Friedman's test

state the hypothesis (step 1)

-hypotheses are part of inferential statistics -types of statistics: descriptive and inferential statistics -types of hypothesis: null hypothesis and alternate hypothesis -in practice, researchers do one of the following: >state the null hypothesis >state the alternative hypothesis >state no hypothesis at all -best practice is to state the null hypothesis, alternative hypothesis, or both

limitations of one-tailed test

-ignores the possibility of an expected difference in the opposite direction -more vulnerable to error if the distribution is not normal

descriptive statistics

-includes collecting, organizing, summarizing, analyzing, and describing/presenting data w/o any inferences abt the population -doesn't tell you if data is significant or not -frequencies and percentages -means and standard deviations -describing a phenomena: how many? how much? BP, HR, BMI, IQ, etc

alpha value

-mostly set at 0.05 -type 1 errors are worse -type 2 is better because you are just missing some info and you overcome that effect

choose sample size (step 3)

-sample size (n) determines: probability distribution to be used and power of the test -z-test required relatively larger sample size (n>30) -t-test is appropriate for small sample size (n<30) when assumptions are met -power is the probability of rejecting the null hypothesis when it is false (correct decision) -*larger sample size typically gives more power to the testing of null hypothesis -studies have a higher risk for Type II errors (accepting null hypothesis when it is actually false) w/ small samples (n<10)*

hypothesis

-statements that describe a proposed relationship btwn 2+ variables -testing this begins with a statement and ends with the decision to either accept of reject it

standard normal distribution

-tells us where the individual scores are located w/in a distribution of scores -here, the scores are expressed as standardized scores (z-scores) Z = (x - x̅ )/s z= standard score x= any 1 score x̅= mean of sample s= standard deviation -raw scores on 2 standardized test cannot be compared bc of different test distributions -ex. 59 on TACL and 60 on PPVT; However, you can compare Z scores

make decision about H0 (step 6)

-test statistic is calculated and compared w/ the critical value -if the test statistic's absolute value exceeds the critical value, H0 is rejected -if the test statistic's absolute value does not exceed the critical value, H0 is accepted -results can sometimes be statistically significant but w/o any clinical relevance -clinical significance is determined by effect size

Alternative Hypothesis (Ha)

-this statement indicates an expected result of a difference btwn the groups -tells you what is better than the other, etc; "yes there is a difference" -most stated

independent samples t-test

-to compare the means of a continuous variable 2 independent samples (ie, 2 different groups of ppl) -do children w ASD have the same lang comprehension skills as children w/o ASD >compare lang comprehension skills across 2 groups, so this is ____ -do patients who receive a new drug tx have better memory skills than those who receive a placebo? >compare memory skills in 2 different groups/conditions, so this is _____

How do I choose a statistical test?

-to test hypotheses about populations, quantitative researchers should choose analytical procedures that match their study's designs -the selection of analytical procedures for testing research hypotheses depends on the level of measurement for the dependent variable (nominal, ordinal, interval, or ratio) as well as the experimental design.

characteristics of normal distribution

-unimodal (ie. 1 mode at the center) -symmetrical: mean and median are the center and 50% of scores are below the center point and 50% of scores are aboce the venter point -scores are continuous fr 1 tail to the other -asymptotic: majority of the scores are in the middle w/ few scores at the edges

accepting/rejecting the null hypothesis

-when the test statistic (ie. observed value) is less than the crititcal value, the N0 is rejected -when the test statistic is greater than the critical value, the null hypothesis is accepted

independent samples t-test assumptions

1) The observations are independent and unrelated 2)Random sampling 3)Normally distributed population 4)Groups should have equal or near equal variances. This is called homogeneity of variance. -Sample size < 30; if it's >30 use normal curve z test -if assumptions of normality or equal variances cannot be met, nonparametric statistic tests (aka distribution-free tests) are performed. -Mann-Whitney U test is performed in this case.

paired samples t test assumptions

1)Two sets of scores must be correlated (related) 2)Random sampling of participants from the population 3)Sampling distribution should be normally distributed -Nonparametric equivalent of t-test related samples designs is Wilcoxon signed-rank test for paired data. -It is appropriate for highly skewed distribution.

process of hypothesis testing

1. state the hypothesis 2. set an acceptable level of risk 3. choose sample size 4. determine critical value 5. compute test statistic 6. make decisions about the null hypothesis (accept or reject)

between group variability

Portion of total variance that is attributable to group membership. It is known as the mean square for effect (MSEFFECT) and is the numerator in the F ratio.

within group variability

This variability is called the mean square for error (MSERROR) and is the denominator in the F ratio.

distribution

a pattern of scores

non-linear

implies curved relationships ex. logarithmic relationships

linear regression

straight line relationship

t-tests computation

student's t= [mean (group 1)-mean (group 2)]/SED -SED: standard error of the difference is a variability or dispersion of scores that is attributable to error. In independent samples, it is estimated fr square root of the pooled variances for the 2 groups -for paired t-test, it is computed as the mean of the differences in scores btwn pairs of participants divided by the SED

statistics

the science of collecting, organizing, summarizing, analyzing, and making inference fr data -2 categories: descriptive and inferential

Null Hypothesis (H0)

this statement indicates no difference/relationship btwn the groups -doesn't tell of a difference -not stated as much bc it doesn't look professional/ makes you less informed -only state this when you cannot make a decision on which tx is better; you're the first and have no idea which is better