Tim F


  • I had an interesting experience when doing a helicopter experiment. Besides many of the factors here, I added paper color, figuring it would be a good way to illustrate a factor that had no effect. Weel, the color didn’t have any main effect, but I got a significant interaction between color and weight of the paper (cardstock vs typping paper).…[Read more]

  • Approximate Solution #1
    The probability of drawing a number, followed by a number 7 larger, followed by a number in between is (about)
    1 * (1/99) * (5/98) = 0.000515
    There are six different orders in which you could draw these three numbers, so multiply by 6
    0.000515*6 = 0.00309[This will be off slightly, because if the first number is 95,…[Read more]

  • I can see a couple of different approaches.On the macro-scale, get the mothly utility bills for each site and compare them to important monthly metrics – sales, payroll, ambient temperature, units shipped … whatever is important to you. Then do some regress calculations. Find out how the electric bill depends on payroll, or how the gas bill…[Read more]

  • Andrea,One caution as you do your study that no one else has mentioned. Ideally, you send out 1037 surveys at random and get back 1037 responses – this will acheive the 99% confidence 4% margin of error you want. If you don’t get 100% response, then ANY results will be questionable. Consider a restaraunt that sets out 100,000 customer…[Read more]

  • Tim F replied to the topic AQLs over 100 in the forum General 15 years, 4 months ago

    Mark,You are simply missing the difference between “defectives” — a part that doesn’t meet spec — and “defects” — any problem with the part.Suppose you have a printed circuit borad with 1000 solder connections. If you have 1 bad connection or 45 bad connections the part is bad. The count for defectives is 1, but the count for defects is…[Read more]

  • Let me add a couple of thoughts to Robert’s excellent post.If the samples were always taken in a similar manner (for example S1 = beginning of roll, S2 = 1/4 of the way along, S3 = 1/2, S4 = 3/4, S5 = end) then you could add S as another 5-level variable in the experiment. You could then look for variations as a function of location along the…[Read more]

  • I’m not sure why *your* MBB likes this rule of thumb ;-) but I can come to a couple of conclusions why this would be a good number. When np > 5, then the normal approximation is pretty good, which is a useful thing for all sorts of calculations. 5, then the normal approximation is pretty good, which is a useful thing for all sorts of calculations.…[Read more]

  • I don’t think there is a “typicaL” value to quote here, because it would depend on too many variables — particularly how tough the defects are to spot and how well-trained & well-motivated the inspectors are.IMHO, the only real answer would be to inspect the inspectors. Have random checks of the products AFTER the inspectors have already done…[Read more]

  • Dude II,Of course, you can only use a normal table if the data is normal to begin with! But I expect Stan knew that ;-)Normality is a common assumption, but by no means necessary. With non-normal data (which is not that rare), then you can only make very broad estimates of the defect rate. A process with a calculated Cpk of 1 or 2 could produce…[Read more]

  • I think there are two different questions lurking here, and they have two very different meanings and two very different answers. I’ll give my $0.02 and see who has any better ideas.QUESTION 1) Is 5 mL an appropriate size sample to draw?
    This questions depends on the test being done. If the test can be adaquately performed on 5 mL and…[Read more]

  • Tim F replied to the topic Hypothesis Testing in the forum General 15 years, 4 months ago

    Julie,The simplest answer would be multiple t-tests. For example, pick some time period you are interested in (for example, the past month or year-to-date). Take all of the data for Mondays and all of the data for Tuesdays and run a t-test on the data. Repeat for M vs W, etc and see if there are any significant differences.The problem is that as…[Read more]

  • One system I did for a presentation was paper helicopters. I only had 1 hr so I pre-made the helicopters, but there are lots of variables you could try:
    * length of blade
    * width of blade
    * length of base
    * width of base
    * cardstock vs typing paper
    * white vs colored paper
    * clockwise vs counterclockwise rotation
    * adding a paper clip
    *…[Read more]

  • Scott,I don’t think there is a really elegant way but here are two options.1) If you are doing a response surface, set the alpha value to “2”. Then you will get 5 evenly spaced values. By default these are -2,-1,0,1,2, but you could change them to anything you want, like 1,2,3,4,5.The downside, is that ALL factors will have five evenly spaced…[Read more]

  • ALMIR,The first thing that jumps out at me is the equation you got:
    Ft of Tubing = – 16.7 + 271 WeightThat means that if you put nothing on the scale, it predicts -16.7 ft of tubing! The constant should be zero if you are going to have any chance of correctly predicting lengths. Look at the three measurements that were farthest off. You cut…[Read more]

  • Tim F replied to the topic What is Rpk in the forum General 15 years, 7 months ago

    According to, it could be:RPK Random Player Killing (gaming)
    RPK Revenue Passenger Kilometre
    RPK Ruchnoi Pulemet Kalashnikov (Soviet Light Machinegun)Any of these sound promising? Are you sure its not Ppk?Tim F

  • Tim F replied to the topic Random Sampling in the forum General 15 years, 7 months ago

    Faceman,That was too simple :-)I was about to grind through something like
    * insert a column of random numbers
    * select all the data
    * sort by the random number
    * select enough rows for that sample you want
    * copy & paste to a new worksheetBut that whole process is already built in to Excel. Cool. Its amazing what some of the software will…[Read more]

  • Tim F replied to the topic Wrong use of P Chart? in the forum General 15 years, 7 months ago

    VPS,Well, I kind of hate to say it, but the process may not be in control! Suppose the process were in control with a long-term average of 95% passing the test. Then you would expect close to 95% in any sample you look at. Perhaps a few more or a few less, but not by much. If you look at subgroups of 1000 pieces, then the standard deviation…[Read more]

  • Tim F replied to the topic Process Z in the forum General 15 years, 7 months ago

    Did you try Minitab help?Process Z
    Process Z is computed by finding the Z-value using standard normal distribution for Average P. The larger the process Z, the better the process performance.Average P
    “Best guess” of the proportion of items in your process that are defective, assuming you have collected enough samples to have a stable…[Read more]

  • Tom,At the simplest level, the control limits are independent of the distribution. They are simply an empirical rule that says anything more that 3 sigma from the center is worth taking a look at. So the control limts are 3 sigma to either side of the center line. If you want to calculate odds, then the normal distribution is handy and it is…[Read more]

  • The space for “Multiple totals” is how much total amount of mixture you want. In the fertilizer example, if you typed in “50 100 200” then Minitab would produce sets of experiments where the total amount of all the fertilizers put together were either 50, 100 or 200. The space for Lower/upper lower limit puts limits on the amount of each…[Read more]

  • Load More