METHODS AND ANALYSIS

A total of 122 samples were measured in this study (Table T1). Samples were obtained as 2.5-cm (1 in) minicores drilled perpendicular to the vertical split face of the rock cores. Samples were taken only from hemicylindrical pieces long enough to ensure that they remained vertically oriented during rotary coring. An upcore orientation mark was scribed on each sample to provide a vertical reference. Samples were spaced at irregular intervals in the core, depending on recovery, unit boundaries, and physical state of core pieces, with the object of collecting at least two to three samples from each lava flow (the range of samples per flow is 1 to 8). Eight flows were not sampled owing to the lack of suitably oriented pieces (Table T1).

Samples were measured with the shipboard direct-current superconducting quantum interference device (DC-SQUID) cryogenic magnetometer onboard JOIDES Resolution. All samples were demagnetized to isolate a characteristic remanent magnetization. Both alternating field (AF) and thermal demagnetization methods were applied to a subset of samples to assess the efficacy of each method. Results from the thermal treatment appeared to better isolate a characteristic remanence than those from AF demagnetization, so thermal demagnetization was used for most samples (Table T1). A total of 101 samples were studied with thermal demagnetization, and the remaining 21 samples were studied with AF demagnetization. Stepwise thermal demagnetization was typically conducted beginning at 150°C and continuing in 50°C steps up to 450°-625°C. On occasion additional measurements were made with 25°C steps above 400° to 450°C. Usually the AF demagnetization proceeded in 5-mT steps from 10-40 or 50 mT and 10-mT steps up to 70 mT, but a few experiments included smaller demagnetization steps.

Demagnetization results from each sample were plotted on an orthogonal vector diagram to aid in finding a characteristic magnetization direction. Using principal component analysis (Kirschvink, 1980), a least-squares line was fit to a straight segment of the demagnetization curve that trended toward the plot origin. The best least-squares magnetization vector was calculated, along with the maximum angle of deflection (MAD) (Kirschvink, 1980), a measure of the scatter in the points constraining the magnetization inversion (Table T1). Characteristic magnetization directions and MAD angles were calculated for each sample using two principal component analysis methods, one with the demagnetization vector tied to the origin and one with the vector free to assume any angle. Although the latter usually produced larger MAD angles, it was preferred because its assumptions were not as restrictive and the results were more consistent. In most cases the difference in direction between the two methods was small (<5°).

Magnetization inclinations from the individual samples were combined to calculate a mean paleolatitude for the site in several steps. We used analysis methods described by Cox and Gordon (1984), which recommend working with colatitude rather than inclination, but are otherwise similar to inclination analysis. First, colatitude values for samples from each individual flow were averaged to determine a flow mean. In several flows, one or two sample measurements were significantly different from other samples in the same flow (typically >2 from the mean of the other samples) and those measurements were not used in calculating the flow average. In two other flows, approximately equal numbers of samples had different signs, requiring reinterpretation of the flow divisions (see below). Flow average colatitude values were tested against those of adjacent flows to see if they were statistically different at the 95% confidence level using the Z-statistic (Kono, 1980). If adjacent means were not statistically distinct, they were combined and the procedure repeated with other adjacent flows. Using this method, a series of statistically distinct flow group means was calculated.

Cox and Gordon (1984) recommend that flow group means be averaged if they represent a time interval less than the coherency time of paleosecular variation so that there is no bias from a time interval that is more heavily sampled. To accomplish this, we judged group means to be serially correlated if they did not show a large change between successive means (~10°) or if a group of colatitudes followed a smooth trend (Cox and Gordon, 1984). Although this procedure is subjective, the result is a more conservative estimate of the number of independent flow mean colatitudes. In addition, the mean colatitude and uncertainty estimates are not greatly affected by differences among the selection of correlated units.

Independent group mean colatitude values (Table T2) and an average colatitude (Table T3) were computed following the methods of Cox and Gordon (1984). This procedure both provides a correction for bias caused by averaging inclination-only data and gives an estimate of the data errors and site mean colatitude. An estimate of the random error is determined from between-group colatitude variations, and an estimate of the colatitude variance produced by secular variation is taken from a model of secular variation data. The final error bounds are calculated including an assumed systematic error of 2° that can be caused by off-vertical tilt of the borehole (a quantity not measured during Leg 191) and correcting for the number of independent group means (Cox and Gordon, 1984).