Measurement/Reliability Flashcards
Discrete Measurment
One Number
Example: One point in an entire cycle of gait.
Continuous Measurement
Same Scale, Multiple different data point taken in a cycle
Example: Kinematics of joint angles moving through a motion
What are the 4 scales of measurement?
Nominal
Ordinal
Interval
Ratio
Nominal Scale
- “dummy variables”; coded 1 or 0
- Lowest level of measurement
- Ex: Yes - No; Male - Female; Injured - Healthy
- Two categorical variables
Ordinal Scale
- Categories order ranked
- “greater than less than relationship”
- good is better than fair which is better than poor
- As many categories as you want
- Ex: Foot strike patterns: forefoot vs rearfoot, high arch vs low arch
Interval Scale
- Can be ranked, each intervals between each rank
- Examples: Temperature
– 1 degree is constant interval
– 30 degrees is “hotter” than 20 degrees - Example: RPE
– 1 is constant interval
– 7 is greater than 3; therefore they are working harder - No zero point
Ratio Scale
- Possess all components of a true number system
- zero, equal intervals
- Example: Peak ground force
– 1000 N is twice as large as 500 N
– 1 N is the constant interval - Example: Running
– I ran 3 miles which is half of 6 miles
Most to Least Level of Measurement
Most
Ratio
Interval
Ordinal
Nominal
Least
What is transforming data? What are the pros and cons? What do we need to consider?
- Changing high level measurements and change them into lower level meaurements
- Example: Strong, Weak
- Pro: Can see the bigger picture (obscures a few data points)
- Con: Lose ability to magnity difference
- Must think about how valid and reliable measures are. Consider if it is error or is true data. The way in which we interpret data changes the conversation (Categories vs Numerical)
Why do we care about the type of data and statistics used to analyze it?
- Influences how you can manipulate the data and what statistics can be used
- Some statistical tests are more “powerful” than others (aka more likely to predict a difference)
- The more powerful the measurment the les people you need to find a difference
Reliability
- Consistency of measurement
- Reproducibility/dependibility of the measurement; free from error
Validity
Accuracy
Tells you if we are getting the correct awnser
What is a “ground truth”?
The measure is what actually happened.
Ex: I ran 2 miles (ground truth); I forgot to stop my watch and now it says I ran 4 (NOT ground truth)
Can something that is reliable be valid? Can we use it in PT practice?
- Yes, if the measurement is reliable/consistent.
- Ex: Scale is always off by 10, you can still get good measurements if you take the 10 into account.
- Consistent but not calibrated
What types of tools are used in PT practice most often?
Those that are reliable and valid!
The wider the data is the ____ reliable
less
Precision gets more important with discrete data. Therefore less precision = ____
less reliable
Decreasing the resolution, eventually leads to..
a much different representation. Not accurate data representation.
Explain what this graph is
Validity: mean of measure and true value
More reliability if the point gets smaller
Measurement Error
- Every measurement includes error
- Measurement is the recorded data point and true score is the “real” or “actual” value under investigation
- Statistical designs can help reduce error, but not remove (repeated measure, within subject)
- The greater the variability between measurements, the lower the reliability is presumed to be
Measurement = True Score + Error
What do we compare when estimating reliability
Expected (True Score) vs What happened (Data w/ Error)
Ratios closer to ____ are reliable
1.0
Primary Contributors to Measurment Errors
Subject
* mood, fatigue, injury type, practice, motivation, knowledge
Testing
* test instructions, multiple testers, tester skill, test environment
Instrumentation
* calibration, measurement drift
Explain this graph in terms of reliability
Trained individuals are more reliable/less variant
Types of Reliability
- Greater reliability increases confidence between repeated measures
- Intra-rater reliability (same person testing)
- Inter-rater reliability (different people testing)
High inter-rater reliability allows for us to…
make easier generalization to the public!
What needs to be considered for generalizability of reliability?
- Inter and intra rated reliability
- Be interpreted with repect to the specific conditions in which they were obtained
- Ex: age, different medical condition, injuries, etc.
Correlation Coefficient
- Describes the extent that one measurement varies with the other – how are different measures related
- Values near one or negative one means there is an association but does NOT explain the extent!
- Ex: Pearson Correlation Coefficent
General rules for r - Correlation Coefficent
- .25 to .5 fair
- .5 to .75 moderate
- Greater than 0.75 good
Can see how correlation coefficent plays into relationship by taking the p and…
- Multiplying it by the percent you want to see change
- EX: As state cigarette tax goes up the % of adult smokers goes down. We have a negative correlation (one goes up, one goes down)
- p = 0.05
- 25% change in state tax results in a change of 25% change in % of people smoking which we got by 0.05 x 0.05 = 0.025 = 25%
What is intraclass correlation coefficent?
- Comparison of two or more repeated measurements
- Provides an estimate of association AND agreement in the absolute magnitude of repeated measures
- Range: 0 to 1.0
- ICC measurements:
– Less than 0.75 = unacceptable
– 0.75-0.90 = good but consider other options
– Greater than 0.90 = excellent and acceptable
Kappa Statistic
- Nominal or Ordinal Data (Categories)
- Gives Rank, the more numbers down the middle the better
Kappa Statistic Values
* Less than 0.2 not different than expected by chance alone
* .2 to .4 minimal agreement
* .4 to .6 moderate
* .6 to .8 substantial
* Greater than 0.8 near perfect (suitable for clinical practice)
