iSixSigma

Data cubes – the death knell for Six Sigma?

Six Sigma – iSixSigma Forums Old Forums General Data cubes – the death knell for Six Sigma?

Viewing 20 posts - 1 through 20 (of 20 total)
  • Author
    Posts
  • #46123

    howe
    Participant

    There’s a big push in our company, and probably many other lately to employ more data cubes. If you don’t know, a data cube is a multidimensional hierarchy of aggregate values. Such a data setup makes finding relationships easier for superficial analysis. Instead of learning about statistical analysis & letting the data lead to conclusions, people are allowing aggregation of data to lead them to conclusions. When I try to look at the data, it’s aggregated and almost always useless from a statistical point of view. When I ask for granular data, it’s unavailable from “the cube”. When I can’t analyze the data, they question Six Sigma’s usefulness… It’s really created a no win situation. What thoughts do you all have on this?

    0
    #152023

    Name of the game
    Participant

    the name of the game in business is: making money. so if the data cubes lead to making money while six sigma leads to conclusions, managers will opt for making money. if six sigma leads to making money while the data cubes lead to conclusions, managers will opt … for the money.  somehow i hear abba in the background singing … .

    0
    #152025

    Name of the game
    Participant

    how could i know about the differences between the highly sophisticated way that six sigma handles data (pie charts, histograms etc.), tqm and all the other efforts that failed including the now legendary data cubes. thanks for the clarification. it’s always so good to learn new things in life from employees who have to mail back to their manager that “there’s not much I can do with it” … you’re up for a real career! good luck, you’ll need it.

    0
    #152024

    howe
    Participant

    Wouldn’t you know it, I just opened up an email from my manager asking for help with analyzing data about which he’s been unable to make statistical sense.  No wonder, it’s all in aggregate form with no rational applied to how it was aggregated!  I have to email him back that there’s not much I can do with it.
    Name, the difference with Six Sigma over TQM and/or vaniilla statistical analysis is that Six Sigma is about making money.  Other efforts in this field weren’t tied to money & failed.  Six Sigma is all about risk/return and how can I make best decisions in order to make money.

    0
    #152046

    Heebeegeebee BB
    Participant

    We are using “the cube” in conjunction with Six Sigma…seems to be a good fit so far.

    0
    #152051

    Robert Butler
    Participant

    I hadn’t heard of data cubes until your post.  A quick search on Google turned up this interesting explanation of how they are built.
    http://www2.cs.uregina.ca/~hamilton/courses/831/notes/dcubes/dcubes.html
      Based on a very quick read it looks like someone somewhere has to make decisions concerning what the cube outputs.  It also looks like someone somewhere has control of the data that goes into a cube.  I’d recommend the e-mail to your boss concerning his/her inability to use the output of the cube should take the form of an educational note. Take what is wanted and explain what is needed to get to where you want to go. Explain why what you have won’t get you there and then perhaps the two of you can find out who controls the data going into the cube and get them to ship you the data needed to solve the problem

    0
    #152060

    Name of the game
    Participant

    can someone please enlighten me why a data cube, i.e. a pivot table is incompatible with six sigma? the cubes are nothing but predetermined, yet flexible summary statistics based on database tables (that’s very different from aggregate data that cannot be traced back to its “granular details”). As such they are descriptive summaries, i.e counts. as a result, the underlying raw data tables can easily be thrown into a data mining tool that allows for the detection of relationships between variables of interest based on for example CHAID analysis. what’s up with this obsession with small sample statistics by six sigma? we’re not on a small farm in rothamstead any more …  

    Account Cube

    ftb
    N

    opb
    (All)

    otb
    (All)

    region
    (All)

    decile
    (All)

    state
    (All)

    division
    Data
    Total

    dv1
    Sum of CustomerCount
    75135

     
    Percent of Total Customers
    61.50%

    dv2
    Sum of CustomerCount
    15749

     
    Percent of Total Customers
    12.89%

    dv3
    Sum of CustomerCount
    31284

     
    Percent of Total Customers
    25.61%

    Total Sum of CustomerCount
    122168

    Total Percent of Total Customers
    100.00%

    0
    #152062

    Haphazard
    Participant

    I had a quick search but not with any rigor, which is why I can’t give you an address, and found a few articles. One of the articles I found used a Splom to show how a Data-cube could find relationships!!!???
    Now I don’t know if a Splom can be considered part of SS or not – SS seems to adopt anything statistical – but if we assume it does, then I don’t see how a Data cube can replace a ‘management philosophy’ or the quest for process excellence?
    Now if someone told me Toyota uses it I might pay more attention …
     

    0
    #152091

    Predictor
    Participant

    While I do think it is possible to do useful work using data cubes / pivot tables, I have serious reservations about them. Given my experience working for one of the largest OLAP vendors for several years, I’d say that most people using these tools are wasting their time. Industry has purchased these tools by the truck-load, and rolled it out to thousands of mid-level corporate personnel, but I don’t think many of them possess the statistical sophistication to use them.My concerns can be summed up by the hypothetical scenario involving hours of “slicing and dicing” the data, followed by an excited dash into the conference room with the words “Two out of three of our customers also use Brand X”. My question is: Does this guy mean, literally, “two out of three”, or “67% of thousands”?This is not an academic issue: Statistical pitfalls are subtle and well-intended but poorly informed analysts.-Will Dwinnell

    0
    #152092

    Hammer
    Participant

    give a hammer to a maniac, and you’ll end up with a bloody mess. that doesn’t mean that hammers are useless or even dangerous …

    0
    #152094

    Haphazard
    Participant

    Agree with your last sentence ..
    I still come across many who beleive correlation is the same as cause and effect.
    I also come across some who don’t even bother to condition data before plotting trend lines, so that the data still contains zeroes from Sunday trading (certain sectors that I won’t mention) and public holidays. It is just a question of garbage in garbage out!

    0
    #152100

    Predictor
    Participant

    Right, so the questions become: 1. Who are these tools appropriate for? …and, more broadly: 2. What percent of people who have and use these tools are qualified to do so?

    0
    #152108

    Steven
    Member

    Here’s another interesting article on it :
    http://www2.cs.uregina.ca/~hamilton/courses/831/notes/dcubes/dcubes.html
    Sounds as though it might be useful in some circumstances.  Six sigma is a mish mash of everything consultants can think of … the more tools, the more training $$$. Data cubes will inevitably be added to the cesspool of SS.
    If you want to focus on quality improvement, use the 7 tools of quality and leave the rest to specialists in your company … if they are relevant.

    0
    #152111

    Hammer
    Participant

    steven, great article. thanks for sharing.

    0
    #152304

    Deanb
    Participant

    TQM used COQ to tie it to money, and just about every improvement program in the last 50 years was tied to financial performance in some way, so saying TQM was not tied to making money while six sigma is is not really correct. The true difference seems to be “who’s” money. In TQM it was the company’s money and the total team shared the credit, mostly nonfinancial. In six sigma there is more emphasis on who gets MORE financial credit, especially the six sigma players. This particular focus on PERSONAL money I do not believe is serving six sigma well, as I have met several CFOs at major companies that are actively de-emphasizing six sigma because of a lack of financial inclusion for the organization as a whole. Data cubes are not the threat to six sigma. Greed and exclusionism are the real threats I see.

    0
    #152311

    GomezAdams
    Participant

    Good point DeanB.

    0
    #152681

    howe
    Participant

    The problem with data cubes is big for a couple of reasons:
    -You lose touch with your data because all you see are aggregates. Data integrity issues are almost a given. You will have problems due to the data’s completeness and the way specific fields are being used.
    -You can’t use the data for statistical analysis involving relationships, e.g. correlation, regression, etc. This is basic college statistics – look up ecological correlation/fallacy.Data is much more complex than the sellers of “business intelligence” solutions would have you believe. Without a clear understanding of the raw data, you’ll surely have failures in decision-making.

    0
    #152688

    Iceberg cubes
    Participant

    Mike,
    I agree with two of your statements: (1) the “top” data are summaries/aggregates and you may lose the details, and (2) that an understanding of the raw data helps. But also keep in mind that you have roll-ups, drill-downs and iceberg cubes. Drill down cubes allow you to disaggregate the data by drilling deeper. This will alleviate the problem that you talked about. It’s not necessarily the “data cube” that is the problem, but what the data cube allows the user to do based on its design.

    0
    #152691

    howe
    Participant

    Ice thanks for the reply, I agree with the second part of your last sentence but this is also what causes the problem with cubes. Also, I don’t see how the drill down will get you to the raw data because it’s just an intersection point and icebergs are aggregates that meet certain conditions, so I don’t see where that helps either.
    I don’t see how you can design a data cube that will support statistical analysis. When you pull data from a cube, it’s no longer going to be continuous data. It will also violate all sorts of assumptions for statistical tools we use because the data will now be dependent, correlated to other values pulled, have sampling error, aggregation bias, & probably more…
    What may be more troubling is that these aren’t designed to answer a business question. Instead people sift through them looking for answers to problems whose existence they never previously contemplated (and obviously not asked for by the customer).

    0
    #152704

    Iceberg
    Participant

    Mike,
    The raw data in data cubes are in the form of tables i,e, counts. As a result, the data are typically nominal or discrete (whichever measurement theory you prefer). A drill down approach allows you to recover the input tables and then run general logit models. If the data are of an ordinal measurement level, you can run multinomial logit models. With iceberg cubes this type of analysis cannot be performed unless you reconstruct the whole series of tables. However, what sales manager will run general or multinomial logit models when they can spend their time on the golf course following sales leads or closing deals???? By contrast, political consultants love this type of exercise because it gives them that kind of special knowledge that Karl Rove raved about before the last election in November :-))))).

    0
Viewing 20 posts - 1 through 20 (of 20 total)

The forum ‘General’ is closed to new topics and replies.