THE DATA COLLECTION AND STATISTICAL ANALYSIS
This practicum was the study of possible correlations between three behavioral (dependent) variables and four predictor (independent) variables. The behavioral variables were represented by the number of timeout incidents per school day, the total time spent in timeout by students each school day, and the number of times students had to be physically restrained during acting out incidents each school day. The predictor variables were represented by the barometric pressure at various times during each school day, the change in barometric pressure during various times each school day, the lunar synodic cycle, and the welfare check cycle. There were three major tasks before the writer. First was an accurate and complete collection of data. For the behavioral variables, this represented an accurate account of daily student behavior. For the predictor variables, it meant collection of barometric, lunar cycle, and welfare check distribution data.
Once all data were collected, the second task required that it be properly sequenced, codified, and entered into a computer file. When all data were appropriately filed, a computer program was run to analyze the data in search of relationships. The Statistical Package for the Social Sciences X was the program used for this purpose, as it is commonly regarded as one of the most appropriate for these calculations. Runs were made on a DEC 20 mainframe computer.
The final task, assuming that relationships were found, was to develop a plan which would include consideration of these relationships in future school programming, and to disseminate these findings to colleagues in other school environments for their consideration and use.
During the 198485 school year this writer had a detailed record kept of all major disciplinary problems through a timeout log. Because the SELC utilizes timeout as punishment for all major disciplinary problems, use of the timeout log gave a good representation of daily disciplinary problems. Survey of the log gives the number of major disciplinary problems each school day, and the severity of the problems as indicated by the amount of time given for each offense. The log also indicates the number of times each day that students had to be physically restrained as an indicator of violent behavior.
The writer obtained a copy of the Climatological Table from the National Oceanographic and Atmospheric Administration, showing barometric pressure for the Hartford, Connecticut area. This was used as the official record of this factor for each day studied. The barometric pressures at seven times for each day: 00:00, 02:00, 04:00, 06:00, 08:00, 10:00, 12:00, and 14:00 were recorded (Appendix A: 41).
Phase of the moon for each date (using 00 for no moon through 28 or 29 for a full moon) was determined from charts in the World Almanac and Book of Facts, and recorded (Appendix B: 47).
In addition, the writer obtained a copy of the dates that welfare checks were mailed in the Hartford area for the period covered by the study. The welfare check distribution schedule (Appendix C: 55) was calculated and prepared for entry using the days from check delivery (i.e. 00= day of check delivery, 07= seven days after check delivery.)
During May and June, 1986 the writer caused all information needed for the study to be tabulated, properly ordered, and prepared for computer entry. Before the timeout data were entered, they were checked for accuracy by a review of all data from the original recording forms. Included in the tabulation of timeout data was the date, and the following information for each date: student attendance, the number of timeouts, the total time of the timeouts, and the number of violent restraints (Appendix D: 56, and Appendix E: 57). While attendance could be treated as an additional behavioral variable, it was used in this study to standardize the timeout data by number of pupils present.
During July all data were entered in the computer in a special file created for that purpose (Appendix A: 41).
During September and October computer runs were made to analyze the data. Frequency distributions were examined to check the validity of the data entry. The Statistical Package for the Social Sciences X (SPSSX) was run to construct scatterplots of predictor variables versus behavioral variable combinations. In order to more carefully determine certain possible relationships, several scatterplot ranges were adjusted (condensed) to make those scatterplots easier to read. At this point, it was noted that data on the behavioral variables had not reflected variation in attendance. The variables were corrected to reflect incidents per 150 attending students, and the scatterplots were reconstructed. These scatterplots should better represent the true variable of interest.
The variables studied consisted of three groupings:
1. Misbehavior (behavioral)
*A. Number of time?outs per day.*B. Average time of time?outs per day.
*C. Number of time?outs requiring physical restraint. adjusted for attendance
2. Weather/Lunar Synodic Cycle (predictor)
A. Barometric pressure each day at 00:00, 04:00, 06:00, 08:00, 10:00, 12:00, and 14:00.B. Amount of barometric pressure change between midnight and 6am, between midnight and noon, and between 8am and 2pm each day (Appendix B: 23 47).
C. Lunar synodic cycle as shown by days after new moon.
3. Welfare check cycle (predictor) as shown by
days after check delivery.
An analysis was conducted of the relationship between each of the three behavior variables and four predictor variables. The analyses were conducted in two stages:
1. Scatterplots for each of the relationships were prepared. This allowed a determination of possible relationships and their shape (linear or nonlinear), and a statistical test for the presence of a linear relationship. The following combinations were examined:
A. the number of timeouts versus each predictor variable
B. the total minutes in timeout versus each predictor variable
C. the number of violent timeouts versus each predictor variable
The Pearson productmoment correlation analysis was
conducted and tested for statistical significance at the
.05 level.
2. Explicit multivariate hypotheses were formulated for each behavioral variable and combination of two predictor variables. These hypotheses were tested with a series of twofactor analyses of variance.
In order to prepare for this analysis, the predictor
variables were first recoded from continuous to categorical variables as
described in Table 1. This process made the predictors more readily understandable.
Basically, those values of the predictor hypothesized to produce behavioral
problems were given a "1" while other values were given a "0". An example
of the resulting analysis is given subsequently.
Table 1
Regrouping of Predictor Variables
 

Values 
Days From
Welfare Check Delivery 


Lunar
Cycle 


Barometric
Pressure 


Barometric
Difference 


In Table 1, the significant categorical variables were defined as: the three days before and after the date of welfare check delivery; the four days around the full moon and around the new moon; barometric pressure below 29.7; and, barometric pressure drop.
Four hypotheses were tested. They stated the existence of a relationship between the predictor variables and the behavioral variables:
1. A drop in the barometric pressure is associated with an increase in behavioral problems.Participant Functions2. Days around a new moon or a full moon are associated with an increase in behavioral problems.
3. Days around the issue of welfare checks are associated with behavioral problems.
4. There is a correlation between paired predictor variables and behavioral problems.
The SELC principal coordinated all activities within the practicum.
He saw that time?out activities were properly logged, and prepared those
statistics in a usable form. He obtained climatological and welfare check
data, and prepared that data for computer entry. When all information was
ready, he caused the data to be entered in the computer file. Once all
data was entered, he coordinated the analysis of the variables by computer.
Due to the nature of this practicum, little monitoring action was required. The writer had to supervise collection of timeout data to assure its completeness and accuracy.
After the first computer run, he had to review the results to assure
that the data had been entered correctly and
plotted in an appropriate way.
Rationale for the Implementation Design
The design of this practicum was dictated by accepted statistical procedures
for the analysis of the relationship between variables.
This author approached the analysis of the data with the firm belief that a strong relationship exists between the behavioral and the predictor variables. In fact, the author was personally convinced that a causal relationship existed. It was with great surprise, therefore, when all of the statistical results indicated only three possible relationships existed between predictor and behavioral variables, at even the .05 level of significance.
The first apparent relationship (Table 2) was found between the Days from Welfare Check and the Minutes in Timeout. While it showed a correlation of .127 at a .045 level of significance, further examination of the scatterplot (Appendix F: 63) and its accompanying statistical data suggested this was attributed to a false positive (Type I) error due to high experimentwise error rate. This interpretation was reached because:
1) A statistically significant result was not found for the same test when the variable was not corrected for attendance;
Table 2
Pearson Correlation Coefficients Showing
Relationship Between Behavioral and Predictor Variables
Behavioral Variables per 150 Students
Variables 



Days from
Welfare Check 



Days from
new Moon 



Barometer
at 00:00 



Barometer
at 04:00 



Barometer
at 06:00 



Barometer
at 08:00 



Barometer
at 10:00 



Barometer
at 12:00 



Barometer
at 14:00 



Barometer
06:0000:00 



Barometer
12:0000:00 



Barometer
08:0014:00 



3) The study ran 36 tests giving rise to the likelihood of a few "false positive" results.
The second apparent relationship (Table 2) was found between the change in barometric pressure from 8am. until 2pm, and the number of violent timeouts. Because this relationship showing a correlation of .138 at a level of significance of .034 seemed to indicate a true relationship, the scatterplot was redesigned (Appendix G: 64) to show a more detailed representation of the relationship through an expanded X axis and the elimination of rounding?off in the plotting of data points. The barometric change results were grouped into three categories. The first category represented the number of violent incidents when the barometric pressure dropped by .10 or more inches. The second category represented these incidents when the pressure dropped between .03 and .10 inches. And the third category represented a drop of less that .03 inches or a rise in pressure (Table 3).
An analysis of Table 3 indicates almost one more
violent timeout on days when barometric pressure dropped more than .10
inches. There is minimal difference when the drop was less, or when there
was a barometric increase. While this is statistically significant the
results raise the question of educational significance. Even if it is known
that significant barometric pressure drops are associated with an increase
in violent behavior, is one additional incident of this behavior sufficient
to warrant major changes in a school's operational plan in anticipation
of this increase. This author feels it is not. Consequently, this finding
does not have educational significance.
Table 3
Analysis of Variance Between Grouped Barometric
Change and Violent Timeouts
Change in Inches 
Days 
Timeouts 









The third apparent relationship was found during a two factor analysis of variance, when one pair of predictor variables showed a significant relationship with one behavior variable. This was found in an analysis comparing the welfare check cycle, the lunar synodic cycle, and the number of timeouts. The cell means of this analysis are shown in Table 4 below.
In Table 4 the predictor variables of the period
within a three day spread around the new and full moon and around a six
day spread of the welfare check delivery schedule, are analyzed in respect
to the behavior variable of the number of timeouts. It is interesting
to note that the highest means occur when either one or the other of the
predictor variables is present, and the lowest means occur when both predictor
variables are either present or absent. Since these results contradict
each other and are contrary to the hypothesized results, it is concluded
that they are a result of a high experimentwise error rate.
Table 4
The Relationship of Synodic Cycle, Welfare
Check Delivery, with Respect to
the Number of Timeouts
SYNODIC CYCLE
WELFARE
CHECK DELIVERY 
