Priscilla's Probability Pullover

Priscilla in her Probability Pullover Our friend Priscilla, aka the Mad Knitter, designed and made a smashing sweater that made good use of her statistical obsessions. This particular method involves simple cables, added on a plain sweater design. Priscilla's particular sweater was a drop-shouldered one from Sidna Farley's "Seamless Sweaters", a good expansion of Elizabeth Zimmermann's work, and reworking of it for looser 80s standards. In this case, it was very easy to continue the cables up the sleeves without having to increase or decrease the number of them. Priscilla liked working bottom up, so she got used to the random pattern before getting into any hairy shaping, and she needs a pattern that can be made in any size. Any similar pattern could be used - circulars are probably best for this. For instance, a custom pattern from the Sweater Machine would do very nicely. Don't get too involved in detailed design at this point, just get the general body size down for a reasonable gauge in stockinette for your yarn.

Think next of what size cables you'd like. Most generally, cables involve 6 knit stitches, separated by 2 purl stitches. But a person can adjust this - 4 knit stitches might be used for finer more detailed-looking cables, 8 for a looser, baggier texture, very stylish in an Italian way right now. And cables can be separated by 1 or 3 purl stitches without any ill effect. Or even none at all, but we don't recommend it for this design because it'd be too confusing. The only constraint here is to stick to the general assumption that cables be done over an even number of knit stitches (an assumption happily challenged elsewhere). And for simplicity's sake we'll also assume crossings all in the same direction. Now let's assume you'd like cables n knit stitches wide, separated by m purl stitches.

The trickiest bit of this method is to get an accurate gauge swatch. Your original pattern design will come with a stockinette gauge. You then need to swatch out what the gauge might be for an ideal finished sweater by making a swatch with at least 6 full cable widths (I'd use 8 for more accuracy), and cabling one single rib each n rows (in the middle somewhere, still for accuracy). In any case, measure the new gauge over 6 cable repeats. Your pattern then needs to be adjusted to allow for not only this gauge rather than the original stockinette one, but if at all possible to give a number of cables divisible by 6. This will no doubt require whipping out the calculator, and working through several iterations of cable size/pattern adjustment till you get something both visually pleasing and likely to fit :-).

Cabled swatch The idea here is that you're going to be cabling on a throw of a die. So, statistically speaking, you'd end up cabling 1/6 of the ribs at any time. But of course theory and reality might not exactly match, so this gauge can only be an approximation. Decide in advance if you'd rather have a tighter or looser sweater and fudge accordingly as you finish designing the sweater. If things get really out of hand as you're knitting, it's permissible to cheat so you don't end up with a weird very baggy or tight stripe in the middle somewhere. But it's recommended that you do so sparingly and with an eye toward randomness:-).

The pattern itself goes like this. Assume again you want to cable over n stitches, separated by m purls. Knit (n-1) rows in regular ribs, (knit n, purl m), and don't worry about a thing. On row n, proceed as follows: throw a die, if you get a 6 cross the cable, if not don't cross and knit as a plain rib. You may also choose some lucky or personally significant number as your crossing number :-), but generally you only want one. Repeat for the next potential cable.

Or in a more formal way:

This will give a perfectly random pattern of crosses, and generally a very pleasing one, much more interesting than you'd get on your own. As you probably know, getting perfectly random distributions is one of the hardest problems to solve, mathematically and with a computer, and mere humans generally stink at it. But one of the points of this sweater is that "random" is not at all "uniform". That is, you don't get an "even" distribution of crosses but rather some pleasing clusters of cables. The whole thing gives the brain a rest from that try-to-discern-a-pattern-from-those-relationships stuff.

A few more pointers can help make a good wearable sweater. Don't cable the edge ribs if you're in/decreasing at the time and don't have a full group to cross. Short row shaping (for bust, back neck or whatever) can easily be disguised and incorporated in this pattern, just turn around at any purl stitches. And because there is a lot of actual ribbing rather than cabling, you get a rather elastic fabric. So closer fit is quite possible to achieve, even with the statistical uncertainities. In addition, Priscilla avoided throwing the die the first four (or so) rows after crossing becuase she didn't want any really tight double- or treble-twisted cables to mess up the works. But as you can see from the above swatch, the latter is not really necessary.

Sleeve detail This sweater itself was made from Opium, a smooth cotton/acrylic/silk blend, in a nice pale green (unfortunately discontinued). The sleeves and body were knit in one piece with Sidna Farley's 'T-top' method. As you can see, there wasn't enough elasticity in this yarn to keep the sleeves from stretching too much :-), but the cables go perfectly smoothly from the body to the sleeves.

This method would naturally apply to all kinds of objects, and it might not be a bad idea to try it in a scarf as an initial introduction for instance. Also, one might use different distributions for shaping - for instance using 2 crossing numbers for a yoke or waist. Or it might work to use more crossing numbers to get a more densely cabled surface, but we don't recommend using 3 or 4 numbers, that'd be more messy looking than interesting because the distribution would be too close to 50/50, and at the higher end having the un-crosses the less frequent could make them look like mistakes rather than artistic omissions. Another alternative would be to use something else than a die for a random number generator - for this swatch, because of being game-challenged, we used cards ace-6 taken from a deck and shuffled, and turned over one by one as we went. This method could allow more flexibility in proportion of cables to ribs, you could simply look for aces in a stack of any size you please.

Warning: as Rose was working on this swatch (from Tekapo, size 8 needles), she became more and more agitated. Her final comment: "I don't mind it being random, but I'd have liked more control". Spoken like her mother's daughter :-). She also points out that it's not really for people who like symmetry. So maybe don't take this on if you aren't ready to let go... Knitting zen?


Baby version Here is a nice baby sweater (named Chaos) using this method. Published in's winter 05 issue.

The designer

Priscilla is currently slaving away in grad school in Baltimore, in epidemiology, and has worked for years as The Data Princess at UCSF, unless you want to call it The Survey Queen. No messy dataset can resist her prying eye. She says about herself: "My mother, a former knitting teacher, taught me to knit when I was eight, I've never followed all of a pattern, usually take the easy way out, was born in the year of the water snake in the belly of the baby boom, am funny, clever, curious and only interested in doing things I can play with (just ask my dissertation chair)."

First published: 10/3/02
Last updated: 4/13/06

All rights reserved. © Fuzzy Galore 2002-2006.