Types of measures:
- Latency
- Frequency
- Rate
- Duration
- Proportion
Proportion:
- The proportion of total time that a behaviour occurred or the proportion of total behaviours that particular behaviour occurred (e.g., if two gorillas spend 30 minutes in rough-and-tumble play out of a total observation interval of 45 minutes, then they spent 30/45 or a proportion of .67 in play). (Note that proportions are not expressed in physical units, such as time, so always be clear in your reports about the total observation intervals or frequencies involved—Martin & Bateson only consider time-based proportions.).
Duration:
- The total amount of time that a single occurrence is manifested during the observation interval.
Rate:
- Frequency per unit time (e.g., Clint pointed 167 times in 18.4 hours of observation, so his pointing rate was 167 points/18.4 hours = 9 points per hour, or 9-1 h).
Frequency:
- The total number of occurrences of a behaviour during the observation interval (note that Martin & Bateson argue that frequency, by itself, is uninformative, so they argue that freq. should always be expressed per unit time, however, there are many circumstances in which it is cumbersome to do so, for example, when the observation interval is identical across all subjects and all experimental or observational conditions).
Latency:
- How long subject takes to respond to a stimulus (e.g., reaction time, or the elapsed time between two successive responses of the same type, etc.).
Scales of measurement - nominal:
- Categorical
- Non-parametric statistics
- E.g., gender, political affiliation, religious affiliation
Scales of measurement - ordinal:
- Ranking
- Non-parametric statistics
- E.g., dominance hierarchies, preferences
Scales of measurement - interval:
- 0 is arbitrary
- Parametric statistics
- E.g., temperature
Scales of measurement - radio-interval:
- Continuous
- Parametric statistics
- E.g., IQ rates, frequencies, reaction times
Recording methods, sampling rules:
· Sampling rules specify which individual is sampled
· Ad libitum - Preferred method for preliminary observations and for rare, but important events.
- Potential bias - tends to miss rare events of short duration and underestimate the contribution of smaller, less conspicuous subjects.
· Focal sampling - A specific individual (or dyad, or family, etc.) is isolated for observation.
- Potential bias - can be large if focal subject seeks privacy for some kinds of behaviours.
· Scan sampling - A number of individuals (typically an entire group) is sampled (typically in rapid succession).
- Potential bias - As for ad libitum sampling, rare events of short duration tend to be underestimated, while conspicuous events are overestimated
· Behaviour sampling (a.k.a. all-occurrences sampling).
- Potential bias - Overestimation of conspicuous events.
Recording rules specify how the behaviour is recorded:
- Time sampling (periodically samples behaviour)
· Instantaneous sampling
· One-zero sampling- Potential bias - Can underestimate rare behaviours of short duration.
2. Continuous recording (records absolute frequencies and durations of behaviour). High fidelity records, but usually means fewer categories of behaviour can be practically coded - Potential bias - Underestimate long-duration behaviours because these are more likely to be truncated by the end of the recording session.
- Potential bias - Can underestimate rare behaviours of short duration.
Coding schemes - principles of measurement:
- There is no such thing as perfect measurement
- Measurements are more or less accurate
· Accuracy is about correct or valid measurement; the degree to which measurements actually capture the phenomena of interest - Measurement are more or less precise
* Precision is about exactitude; the degree to which measurements are reliable or replicable
- Measurements are more or less accurate
Research question 1 - is the baby distressed?:
- Limitations:
- Misses change in intensity of crying
- Uses arbitrary clip boundaries, instead of relevant events to define observational intervals
- Scheme does not capture differences in the intensity of baby’s responses to different social contexts
- Same for research question 2
Intra and inter-observer reliability - preliminary considerations:
- Intra-observer reliability means the same observer coded the same behavioural record at different times (obviously, only possible with video, audio, or transcript records)
- Inter-observer reliability means that different observers independently coded the same behavioural record either at the same time or at different times
- No strict standard currently exists for how much of a behavioural record to independently code, for purposes of reliability, but typically it is around 15% to 20% of the record
Inter-observer reliability - consensus vs consistency:
- Consensus estimates are based on the assumption that two or more coders can come to exact agreement (percent agreement, kappa).
- Typically used on nominal data, where different codes represent qualitatively distinct ideas.
· Consistency estimates are based on the assumption that it is unnecessary for two or more coders to interpret a scale identically, but that the coders will consistently classify phenomena with their understanding of the scale (Pearson’s r, Spearman’s rho, Cronbach’s alpha). - Typically used on ordinal-to-continuous scales
- Typically used on nominal data, where different codes represent qualitatively distinct ideas.
Inter-observer reliability - types of measures in common use:
· Consensus
- Percent agreement - Does not correct for agreement by random chance
- Cohen’s kappa - Proportion of agreement after corrections for agreement by random chance
· Consistency
- Correlation coefficient - Does not take into account variance between coders
- Cronbach’s α - Corrects for variance between coders, can estimate reliability for more than two coders.
