Announcement

Collapse
No announcement yet.

Samples: Help me do my job!

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

    Samples: Help me do my job!

    I've got 1,000,000 widgets split into 60,000 groups that are similar sized but not equal. Each group is broadly similar but has four or five characteristics that differ in a reasonably narrow range. How many do I need to sample to get a representative view of the whole? Is that number different if the number of groups is higher (say 250k)?

    I don't necessarily need definitive answers, I'm just hoping to be pointed in the right direction to do some background reading on the subject.

    #2
    Samples: Help me do my job!

    Do the characteristics overlap, or are they mutually exclusive?

    Comment


      #3
      Samples: Help me do my job!

      The characteristics themselves are mutually exclusive. Age, type and location would be three of them for instance.

      Comment


        #4
        Samples: Help me do my job!

        Why are you profiling the characteristics of midgets?

        Comment


          #5
          Samples: Help me do my job!

          Oh, sorry.

          Comment


            #6
            Samples: Help me do my job!

            I guess the simple answer is that you sample as many as is manageable, as the greater the size of your sample, the better its representation of the population. I was wondering about the characteristics from the point of view of looking at a stratified sample but with a dataset of one million and a large sample I would have thought that simple random should be ok.

            Comment


              #7
              Samples: Help me do my job!

              There is a sample size calculator here:
              http://www.surveysystem.com/sscalc.htm

              However the practical use of this is for constructing a confidence interval of a population characteristic expressed numerically. That may or may not be of specific use to you but you might find that plugging some numbers in to the box gives you a feel for possible sample sizes.

              Comment


                #8
                Samples: Help me do my job!

                I've no idea what any of this means.
                I'm somewhat disappointed to find out that you're not a carpet salesman.

                Comment


                  #9
                  Samples: Help me do my job!

                  Thanks kev, that is very useful and helpful. It tells me that a sample of 600 is needed for a confidence interval of 4 at a confidence level of 95%. The big caveat is that it assumes a randomness to the sampling.

                  However, the selection of groups in the survey that I am evaluating is not random. I have a survey of 700 widgets but only based on 4 groups. The groups are biased towards a particular age profile. A random selection would have a confidence interval of 3.7. How do I figure out the effect of non-random sampling?

                  Comment


                    #10
                    Samples: Help me do my job!

                    It would be hard to figure out the effects of it. It's important to know if you are 'missing' any age groups, or whether they are just under-represented, in your sample. If they are absent, then there is a problem with the sample. If they are under-represented, you could multiply the different age groups by a weighting factor to give them approximately the same representation in your sample as in the population.

                    Simple e.g. A population has:

                    35% over 65s
                    25% age 50-65
                    20% age 35-49
                    20% age under 35

                    Say in your sample of 1000 people you have ended up with:

                    400
                    150
                    250
                    200

                    What you can do is weight the results of the first group by 35/40 (the desired proportion divided by the sampled proportion) so that their significance is reduced to what you want. The second group are under-represented in your sample (15% when you wanted 25%) so you can multiply their results by 25/15, and so on.

                    It's a bit agricultural, and you'd have to watch out for the obvious hazards involved with giving a very small sample group a large weighting.

                    Comment


                      #11
                      Samples: Help me do my job!

                      Widgets, eh? We deal with them a lot in my trade. What's your marketing plan look like?

                      Comment


                        #12
                        Samples: Help me do my job!

                        Aren't they the things in the bottom of beer tins?

                        Comment


                          #13
                          Samples: Help me do my job!

                          Yeah, I've seen them in Guinness. Do others use them?

                          Comment


                            #14
                            Samples: Help me do my job!

                            Ask Jack Dee.

                            Comment


                              #15
                              Samples: Help me do my job!

                              Yup....don't get it.

                              Comment


                                #16
                                Samples: Help me do my job!

                                Comment


                                  #17
                                  Samples: Help me do my job!

                                  WOM wrote: Yup....don't get it.
                                  You don't get a reference to a UK only telly ad from fifteen+ years ago? Pathetic...

                                  Comment


                                    #18
                                    Samples: Help me do my job!

                                    Someone nicked my issue of Strategy that week.

                                    Comment


                                      #19
                                      Samples: Help me do my job!

                                      The widget. Great invention, that.

                                      I must get me some Guinness in.

                                      Comment

                                      Working...
                                      X