
Methods  Page 4 

Introduction
 
Results

Discussion

References

Added figures

Figures
 
(Continued from previous page) Most of the recordings from viable single cells (including all from the hypothalamus and all related to affective defense) were initially negative in waveform and were presumably obtained from cell bodies (Fig. 2). Ten recordings from the superior colliculus and reticular formation were purely positive, however, and they may have come from axons (7). It may be noted that less than 5% of the cells with normal waveform were injured according to other criteria, while most of the cells with notching or positivenegative waveforms either showed other signs of injury or else were unresponsive to any manipulation while all other cells from the same region were responsive. It was possible to record from twothirds of the single cells encountered in the study before, during, and after at least one instance of affective defense without injury, pronounced changes in spike amplitude, or loss of recording contact with the cell. Most cells were held for ten or fifteen minutes at least while a trial of affective defense, a full set of control manipulations, and a second trial of affective defense were conducted. Some cells were held for periods of up to two hours and during as many as eleven. displays of affective defense. On some occasions recording was maintained from a cell while the recording cat was rolling over and wrestling with the attacking cat. If the electrode was left in the same position for more than several hours, however, recording could no longer be maintained during movement by the cat. During movement spike amplitude would systematically decrease and then return to normal after cessation of movement. Statistical treatment of data Two kinds of measures were made of the response of each cell to each manipulation of the cat: a onesecond measure of the number of spikes during the second coinciding with the beginning of the event marker, and a fivesecond measure of the number of spikes during five seconds beginning with the event marker. For affective defense, however, it was necessary to begin the onesecond and fivesecond measures a second prior to the event marker in order to take into consideration the reaction time of the experimenter. For cells which were inhibited but which had low baseline firing rates it was necessary, in a few cases, to use a tensecond measure instead of the fivesecond measure or a two second measure instead of a onesecond measure for lifting of the cat in order to achieve statistical significance. The baseline firing rate for each cell was determined by counting the number of spikes per second in each of the ten seconds of each baseline period preceding each attack trial and control manipulations trial. The data from all baseline periods of a particular cell were pooled, the form of their distribution was approximately by a Poisson or normal distribution (see Fig. 3) and the goodness of fit was tested by a X² test. Of the baseline distributions of the 9.1 cells, 67 were accepted as Poisson at the 0.10 confidence level, nine others which were rejected as Poisson were accepted as normal at 0.10 confidence level and nine had baseline firing rates of zero. Of the remaining 10 baseline distributions, four which always fired in pairs or bursts of spikes could be fitted by a Poisson distribution if each pair or burst was treated as a single event and the other six distributions all had modal values at zero which made them appear more like Poisson than normal distributions, so they were treated as the former. It should be noted that it usually made little difference in the final analysis whether Poisson or normal distributions were used to fit the data. The actual changes in firing rate in response to manipulations of the cat were generally far greater than the differences between values predicted by Poisson as opposed to normal distributions. For cells with Poisson baseline distributions, the probabilities for onesecond measures were obtained directly from published tables of the Poisson distribution (10). Poisson probabilities for fivesecond measures could also be obtained directly from the Molina tables by using the sum of the fivesecond measures and referring them to the Poisson distribution expected for a mean five times that of the cell's baseline distribution. This follows from the theorem that "the sum of any finite number of independent Poissonian variates is itself a Poissonian variate, with mean equal to the sums of the means of the separate variates" (12). For cells with normal baseline distributions, the probabilities for onesecond measures were obtained by converting each onesecond measure to a standard score based on the mean and standard deviation of the baseline distribution. Probabilities for the fivesecond measures were also obtained by standard scores, comparing the mean of the fivesecond measure to the normal distribution expected with a mean equal to that of the baseline distribution, but with a variance only onefifth as great. This follows from the theorem for standard error of the mean of the normal distribution in which the variance of a distribution of sample means is reduced in proportion to the number of scores used to determine each sample mean. Twosecond and tensecond measures were treated in the same manner as the fivesecond measure. A cell was considered to have been facilitated (or inhibited) by a manipulation if any of the above measures were consistently above or below the baseline distribution at less than the 0.02 probability level.
