1/16
Looks like no tags are added yet.
Developing a Hypothesis
Hypotheses: a tentative explanation or prediction about the outcome of a research investigation
Null (Ho): statement of equality or no relationship between variables (statistical purposes)
Alternative (Ha): statement of relationship between variables (directional or not) = typically
Elements of a Good Hypothesis:
Stated in a declarative form (statement, NOT a question)
Identifies a specific expected relationship between variables (predicts direction)
Reflects and is based on theory and related, already existing, literature
Concise, specific, and testable (falsifiable)
The hypothesis has to be made BEFORE the research, it is a prediction + based on question
Null: mean 1 = mean 2
Alternative: 1 < 2 or 1 > 2 or 1 does not equal 2
Priori hypothesis is done BEFORE the study is conducted
Directional vs. Non-directional Hypotheses:
Non-directional predicts that a difference or relationship exists, but doesn’t specify what direction
Null: no difference in fat loss between aerobic and anaerobic exercise
Alternative: fat loss differs between aerobic and anaerobic exercise
Directional: predicts that a difference exists and which group will have higher/lower value
Null: no difference in fat loss between aerobic and anaerobic exercise
Alternative: aerobic exercise will have a greater difference on fat loss than anaerobic
Testing Hypothesis
Procedure used to determine if the finding of an investigation are “real” and not due to chance (statistical significance) → are you seeing actual differences or is it random
is the difference significant enough to determine you can REJECT the null
Four steps:
State the hypothesis
a tentative explanation or prediction of the outcome of a research investigation
Null: directly assessed via hypothesis testing VS. alternative: not directly assessed
Select the criterion / criteria for rejecting the null (to make it consistent)
Alpha: the level of uncertainty a research is okay with
usually .05, meaning comfortable with 5% odds that results are due to chance
pre-determined criteria for level of uncertainty going into the study
below = reject the null hypothesis VS. above = fail to reject the null hypothesis
Compute the test statistic
collect and analyze the data → test statistic and p-value
Reject or fail to reject the hypothesis
you do not accept or reject the alternative hypothesis due to uncertainty because statistics do not prove anything
our p-value may be less than alpha (p < 0.05) → REJECT = chance is very small
our p-value may be greater or equal to alpha (p > 0.05) → FAIL TO REJECT = chance >5%
p-value: how compatible the observed results are with the assumption null is true
p < .05 = statistically significant → evidence to reject the null hypothesis
p > .05 = not statistically significant → fail to reject the null hypothesis
Inferential Statistics: most research in EXSS involves finding the difference between 2 things
between group differences: injury prevention improved movement quality in youth soccer players compared to regular warm-up → determine if difference is by chance or intervention
differences across time w/ same group : following rehab, concussion symptoms decreased between 7 and 21 days post-concussion
Goal: measurements obtained from a sample (the statistic) are reasonable estimates of the same values in the population (parameter)
differences you see are NOT due to chance or error
You are inferring something about the population using a sample
Sample Error: where differences in result are due to chance or random effects
sample error is always present to some degree
miscalibration, social desirability, bias, lack of understanding, etc.
Effect Size: the measure of HOW different two groups are from one another (magnitude)
sample size is NOT included
Type I and Type II error
Type 1 Error (alpha): rejecting the null with it was true → false positive
hallucinating a difference (saying there is one when there is not)
reduce by decreasing alpha
Type 2 Error (beta): failing to reject the null when it was false → false negative
missing a difference (saying there is no difference when there is one)
reduce by increasing alpha and/or sample size
Consequences of Type 1 and Type 2 Error:
Type 1: saying a medication makes a difference when it does not → expensive & unsafe
Type 2: saying a medication doesn’t work when it does → missed opportunity
Type 1 is considered to have more serious consequences → giving misinformation
Statistical Power
Power = 1 - probability of type 2 error (chance of missing an effect) → finding true effect
increase sample size = increase power
but increasing too much can create too much power with little to know meaningful difference
Goal: about 80% power (.2 beta) = 80% chance of real effect
alpha and beta are inverse
Increase statistical power:
larger sample size, larger true effect size, lower noise, larger alpha
larger sample size = more information and reduces noise
Too much power being a problem:
a very large sample size can make insignificant difference statistically significant
why priori hypothesis are good → estimate appropriate sample size before study
6 Steps to formulate a strong hypothesis
Ask a question
do some preliminary research → base research off of existing knowledge
formulate hypothesis → clear and concise statement
refine hypothesis → specific and testable + PICO
phase hypothesis in 3 ways = If…then…, correlation/effect, comparison
write a null hypothesis → for statistical research
Chi-square tests (Goodness of fit test)
Chi-square of goodness of fit test allows for testing if the distribution of frequencies is different from what is expected → rates or frequencies in categories → categorical data (nominal)
observed: actual number of occurrences in study
expected: the expected number of occurrences based on past literature (if null is true)
(O-E)² / E
Chi-square goodness of fit: examines one categorical variable (observed vs. expected)
ex: are injury types distributed equally among athletes
Chi-square test of independence: examines two categorical variables and whether the distribution of one variable is associated with the other
ex: is injury type associated with the sport the gender of the athlete
Degrees of Freedom (T-Tests) + Critical Values + Tailed
Degrees of Freedom: the number of values in a calculation that are free to vary
if you have 5 numbers and know the mean is 5, once you know 4 of the values you can determine the 5th value, therefore df = 4
Critical value: a threshold used to determine whether a test statistic is sufficiently large to be considered statistically significant
Critical value is determined based on degrees of freedom
if chi-square is greater than critical value → reject the null / less than → fail to reject
One-tailed vs. Two-tailed: 1= directional (a= .05), 2= non-directional (a= .025)
one-tailed: does an active rehab improve return to play times
two-tailed: does a new strength training program change VO2 max in collegiate athletes compared to a control group
T-tests
T-tests: determine whether there is a statistically significant difference between means
Compare means between groups or time points
use continuous numerical data → interval or ratio data
assumes a normal distribution of the dependent variable
the larger the sample, the more a t-distribution approaches the z-distribution
Are these 2 groups different enough that it is unlikely to be random noise
Single Sample t-tests: 1 group vs. population (df = n-1)
Allows to test the difference between a distribution of a continuous variable and a interest
ex: are athlete’s current beach press results higher than the historical standard?
one sample mean against one known value
Paired samples / dependent t-test: 1 group, two time points (pre-test and post-test) (df = n-1)
Test to see if there is a significant difference between the two scores from related samples
good for: rehab or recovery + performance improvement
ex: compared to last year’s bench press, are the athlete’s stronger this year?
pre vs post training
Independent t-test: 2 groups (df = n-2)
Test to see if there is a significant difference between the means of 2 independent groups
ex: are ACC athletes stronger than SEC athletes on bench press?
t-test vs. z-test
a t-score uses a t-distribution which accounts for the uncertainty due to estimating population SD from the sample
use t-score and degrees of freedom to determine p-value
z-test assumes that the population SD is known, but most research it is unknown, we only have sample SD → so we use t-test
bessel’s correction: n-1 to help avoid underestimating the true variability in the population
Using a t-test for 3 categories accumulates 5% chance of chance to 15% = increased false +
ANOVA = Analysis of Variance
Comparison among 3(+) groups = measures the different types of variance that appear in the data and then explains the source of each variance
Between subjects ANOVA → 3 different groups or 3 different levels
Repeated measures ANOVA → 3+ time points
Is the variation between the groups big compared to the variation inside them
Total variability: the difference of each score from the overall grand mean
Between: the effect of the IV
amount of variation attributed to differences between groups
the difference of each group’s mean from the overall or grand mean
Within: individual differences
amount of variation attributed to differences from person to person
the difference of the individual scores from the mean group
F-ratio: (between x treatment)/within = IV
F = 1: the amount of variability within groups is equal to the amount between → NOT
F < 1: the between groups is < within groups → NOT significant
F > 1: the between groups is > within groups → more LIKELY significant
Types of ANOVA
One-way ANOVA: one dependent variable + one independent variable + 3 groups
Repeated Measures ANOVA: compares the difference in the DV for a single IV when you have multiple data points from a related sample
same group at 3 or more time points
before → during → after but SAME people for all
everyone acting as their own control
Factorial ANOVA: tests the effect of two or more factors (IVs) with 2 or more levels on one dependent variable + how those factors interact
factor = independent variable
Mixed ANOVA: 1 between subjects and 1 within subjects
ex: 1) class_pre, 2) control_pre, 3) class_8weeks, 4) control_8weeks
Post Hoc Analyses
Tell you WHERE the difference is
Tukey HSD: comparing groups against each other → Q-distribution
equal sized groups → balanced designs → adjusts p-value
common for one way
less conservative
Bonferroni: divides alpha by number of comparisons
more conservative → small number of planned comparisons
better for unequal groups
stricter, but more flexible than tukey
Degrees of Freedom ANOVA
between: number of groups - 1
within: participants - number of groups
total: participants - 1
F(2,27)
2 = between
27 = within
Effect Size (eta² = n²)
How much of the variability is explained by the effect → .25 = 25% explained by that variability
small = .01
medium = .06
large = .14
Partial n² → factorial or repeated → isolates each factor’s unique contribution
Levene’s test of homogeneity: homogeneity of variance → are the variances roughly equal
not significant (p > .05) → assumption is met
significant (p < .05) → variances differ → ANOVA becomes less reliable
Correlation
Correlation: a measure of strength and direction of a linear relationship between two variables
as one changes, does the other in a predictable way
Typically measured with a correlation coeffcient:
Pearson r: assumes normal distribution with interval or ratio data → standardized
-1, 0, 1: closer to 1 = stronger the relationship → neg = inverse vs. pos = direct
continuous variables w/ no strong outliers
assumes normality
measures a linear relationship
Spearman (rho): nonparametric procedure with ordinal, interval, or ratio data (flexible)
ordinal or nonparametric data
does not require normality
uses ranked values and can assess monotonic relationships
Are the dots close enough to the line to indicate significant relationship between variables
Regression
Specify one variable as the predictor (X) and one as the outcome (Y)
X = the independent variable / predictor
Y = the dependent variable / outcome
b = unstandardized coefficients / slope
a = intercept / Y when X is 0
Is the line itself significant (slope)