Section II: Behavioral Observation Methods

Lecture Plan

I.  Introduction to Methodology
This unit teaches about important aspects of how behavioral observation data
are structured  for  analysis. It deals with one simple idea, patterning, but it requires
thinking about the methods involved. "Patterns" are repetitions of events in time.
There are two ways of talking about repetitions in time:
(a) the total number of events per unit of time, or base rate, and
Response events, SR  units as they occurred over time.

For example: How many times do you touch your face in 10 minutes?
(Total number of face touches.)
How many of those face-touch times occur when I call on you?
Now  the unit is--
[ Stimulus: Calls on;   Response: Face Touches]

If I know your touch rate is 4 times per minute, whether or not I call on you,
I can also ask whether you touch more or less (than EXPECTED), given
that I called on you. In terms of probability:

we would determine that your rate of face touches is, say .25. (If I secretly
looked in on you, you are likely to touch your face 25% of the time.)
That is your base rate of face touches. It is also the unconditional probabilty
(it just happens, it is not conditional upon somehting else having happened first).
Now I can also calculate how many of those face touches happened when
I call on you.

Possibilities: all of them occur ONLY when I call you,  100%, or
50% occur when I call on you.
3% occur, and so on, all the way to 0% occur when
I call on you.

So  you see,  the pattern  [calls on -- face touches] may
be more or less likely than just knowing  you are a 25% face-toucher.
But don't be confused: your total face-touch "pie" is still 25%.
We are only asking whether the events represented by the 25% of
the time occur more than expected ( i.e, just  25%)  when something
else has just  happened.

That is, the conditional probabilit is that  probability of  some event,
given some other prior (conditional) event.

For example, H+ given W- is NOT the same as the unconditional
probability of H+. He may be  "more"likely to be  positive when she is
negative than we expected, i.e., more positive than  just knowing his
base rate of being positive (H+).

II.  Understanding methods for describing patterns of interactions

Data stream = the actual depiction of the events according to the code-book.
Base rates = the observed rate at which something happens over time,
as in total events  divided by total time

E.g., 33 positives in 10 minutes = 33/10 = 3.3 per minute

Sequential analysis = method for describing unfolding patterns
Retains the sequence (or order) of events, i.e., which
came first, etc.

E.g., H+ W- H- W- H+ W- W+ ..................

Example:  Base rates (from this sequence):
2 H+, 1 H-, 1 W+, 3 W-  (in x minutes)

Sequential analysis (Lag-1):

 Response
 Stimulus:: H+ H- W+ W- Total H+ 2 2 H- 1 1 2 W+ (1)* W- 1 1 1 3
* Series ends with W+ so there is no "response" to it; not counted in Total
Explanation:
In the sequence or line of codes, above, we look for each code as a
stimulus (leading) and each code as a response (following). Each
code serves in both roles. We now move across the codes with a
WINDOW of size one, which means that when we look at the
first code (H+) above we ask what was the VERY NEXT code
response: it happened to be W-. So, in our summary table above
we note 1 for the combination H+ W- . But then it happened again,
so we need to tab another 1.We go through the entire chain of
adjacent codes. If the window is only open for one code then
we look at each adjacent code. Convince yourself that is what
I have done in the summary table. (Note: the summary table
lists all possible combinations of stimulus and response codes.
Some are blank because why?)

Now think back to the Systems pictureI showed you of M,F,C in Section I.
We said that by looking at the frequency with which various combinations
occurred we could tell something about the family.

This is especially true here too. Let's say that for the PATTERN H+ W-
there were 10 counts of this event sequence in the chain: what would
immediately by wife doing something negative.)

What if I made the window size 2, or 3 etc.? What would you say
if the H+W- pattern occurred 10 times when viewed through a
window as large as 4? That means that after each H+ we get a
W- only after three other responses intervened, H+ [ 1, 2, 3] W-
. What would you make of the timing of this pattern for the couple?

III.  Defining Forms of Reciprocity

Reciprocity refers to an exchange of events: A does something which
is then followed by B doing something. The measurement of A-->B
sequences is the primary objective of this unit on behavioral observation.

There are two forms of reciprocity that we need to be concerned with:

Base Rate and Sequential or Serial reciprocity

Base rate always refers here to the total output of some behavior,
regardless of order effects. All the "smiles" a person emits in a period
of time is what we refer to as his/her "base rate of smiles"

Base rates, in one sense, serve to describe a "personality" of an
interaction. If we know that H has a base rate of smiles that is .45,
and W has a rate of .68, we could say that this couples puts out a
combined total of more than 50% of smiling behavior (i.e., ~.5
of all their behavior coded in fixed time will be smiles). This is
a happy couple!

Base rate reciprocity
We can define the likelihood (or probability) that any given observed
behavior will be of type X (e.g., smiles) by simply using X divided by
all behavior, as above. The only new idea here is that p (the probability)
can be used just like a score. If each spouse has a p value for their
own version of behavior X, we have two p values or scores. Now,
over couples, we can compute the correlation (r) between these p
values, one for H and one for W as:

p(H+) or  p(W+) Husband positive,  or Wife positive

p(H-) or  p(W-) Husband negative, or  Wife negative

If the correlations are high we would say that there is reciprocity
between Hs and Ws, such that the more one puts out X behavior,
the more the other reciprocates. In these instances, we are describing
positive and negative behaviors.

Sequential or serial reciprocity
Conditional probability forms the basis for the scores for each
spouse. Conditional probability refers simply to a limitation of
events, such that a final event is dependent upon (or conditional
upon) something else having been true. This is neat for marriage
work because we actually want to see whether one person's behavior
is dependent upon the other's behavior! We can do this simply
using the notion of conditional probabilities. Just as we defined
a score above as p(H+) we can also define a conditional
score which says the probability of H+ given W+ is such
and such a number. The higher this number the more likely it
is that H will be + WHEN W is also +. Remember, H can be +
whenever he wants (limited by his own base rate of H+), but
if he makes his H+s only when W is (or has been) +, then
clearly he is dependent on her behavior. Conditional probabilities
are written according to the following codes:

p(H+ | W+) or (W-| H-)

The vertical line stands for "given". You would read these as "Husband
positive given wife positive", etc. What we would like to know is
whether the husband's behavior is really tied to his wife's behavior,
as in the notations in the following examples

p(H+ | W+) > p( H+) or p(H- | W-)>( H-)

But --we said before that the conditional probability cannot be greater
than the base rate (unconditional) probability. He cannot be more
likely to do something when she does something than how much he
actually does something (i.e., his base rate).
So the ">" or the "<" signs are relative to something else, namely,
what we would have expected him to do naturally (unconditionally).
We need not concern ourselves with how this is calculated. but it
follows the usual Observed minus Expected formulas.

Now putting this together with reciprocity we can say that two
sets of behavior patterns are reciprocal (or linked) if there is a
high correlation between spouses on these very patterns, or

p(H+ | W+)  .  p(W+ | H+)

p(H- | W-)   .    p(W- | H-)

In words, this says the correlation between the pattern [Husband
positive, given Wife positive] and the pattern of [Wife positive,
given husband positive] is such and so. Like before, in the case
reciprocal patterns. See for example, that there could be a husband -
given wife -   pattern that was NOT reciprocated wife- given husband-.
These are separate possibilities, and I have only illustrated two of them.

Negative Affect Reciprocity within Person:

On two different occasions, H and W discuss a problem (e.g., Sex),
and then what would be fun to  do as a couple. Their behavior is
coded as + or - on each occasion. The p values are determined
and the correlations between sex and fun discussions are run, with
these results:

Discussion Topic :                          Sex

 Fun H+ H- W+ W- H+ 27 H- 36 -22 23 W+ W- 29 -32 34
First note the correlations (bold are significant; empty cells rs too small) between
time 1 and time 2 within each person (.36 and .34). This means that each person's
base rate remains relatively stable (?) across the situations; if you are negative in
one you are also negative in the other. (Not huge correlations, but a significant ones.)
But note, there are not these cross situation r's for positives! This suggest that
negative affect is reciprocal within persons (i.e., consistent). The two smaller
r's (.23 and .29) between spouses indicate a tendency (although not significant)
for there to be reciprocity between them for negative (i.e., if one is negative
time 1 the other is negative time 2).

Same study below, (two occasions) but now the rs are between the sequential
patterns discussed above. Here the force of negative affect reciprocity is very clear
The rs tell us that a pattern of (say) wife negative given husband negative during
the time 1 discussion is reciprocated in the time 2 discussion (r = .39). But even
more striking are the H-|W- W-|H- (r = .62) and the W-|H- H-|W- (r = .54)
reciprocal patterns. These indicate that regardless of who starts being negative
the other person is more likely to be negative. [These data are from as study
reported by Gottman , 1980, and are used here to illustrate the methodology.]

Discussion:                                                  Sex
 Fun W+|H+ W-|H- H+|W+ H-|W- W+|H+ W-|H- 39 62 H+|W+ H-|W- 54 46

See "Further explanation"

III. Coding Illustrations

A. Coding Options -- different ways of portraying methods for observing
interaction data

B. Data Stream -- options for clustering data (i.e., discrete code, sequences)

C.  Examples of cumulative point graphs

These show the trend within a discussion,whether increasingly positive
(regulated) or  increasingly negative (non regulated).  Using the trend
(point graphs) as a predictor we  can then decide whether wehave types
of couples (i.e., satisfied or  divorce prone, etc.).  Positive and negative
codes are assigned numerical weights, and the algebraic sum of the
weights are plotted for every talk turn. E.g., +5 -3 +2-1 =  +3. If the
discussion gets more  and positive (i.e., +'s outnumber -'s) the trend
is upward; scores are added to the immediately  previous score
(= cumulative points).

Continue to next page
Behavioral Observation