+ Reply to Thread
Results 1 to 7 of 7

Thread: (SPSS) External validation for a linear regression model

  1. #1
    Points: 119, Level: 2
    Level completed: 38%, Points required for next Level: 31

    Location
    Glasgow
    Posts
    4
    Thanks
    2
    Thanked 0 Times in 0 Posts

    (SPSS) External validation for a linear regression model




    Hello,
    I am investigating relationships of continuous variables and I am aiming at getting a model that can be generalized for the study. My DV is indoor pollutant and selected IVs are behaviors on ventilation devices and relative humidity. I did Pearson correlation first and got the indication of significance on those IVs. I did curve fit on those IVs too prior to conduct the multiple linear regression to find if those IVs are not linearly correlated. Having checked that linear assumptions on them all are valid then I did regression. So using IVs that indicated p-value of < 0.05 (from Pearson correlation), the linear regression model was built. Now, I have another set of raw data from other project that I could use for external validation (same variables as per model). I checked this separate set of data and unfortunately, it did not give me the same outcome on significance from Pearson correlation. Anyway, is this important at all prior to validate the model? Also as this is not cross validation (so theory on random select case is not suitable), do you guys mind sharing with me on how to conduct this on SPSS please? I checked theory on comparing Rsquare back to the model, but how to do it using this external data on SPSS?

    As a background information, I am at the beginner level on SPSS, so I don't have any knowledge on programming to use the free R software.

    Thank you in advance for your kind attention.
    With all my best wishes to you all in the New Year 2015 too.

  2. #2
    Omega Contributor
    Points: 38,303, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,993
    Thanks
    397
    Thanked 1,185 Times in 1,146 Posts

    Re: (SPSS) External validation for a linear regression model

    The linear assumption is on the error terms in the regression model.

    Can you create a table of descriptive stats to compare the samples to see where they differ. Also how might these samples differ from the population.
    How where the two samples collected (convenience samples) ?

    Perhaps you need to establish inclusion and exclusion criteria to make a model generalizable!
    Stop cowardice, ban guns!

  3. The Following User Says Thank You to hlsmith For This Useful Post:

    wid (01-03-2015)

  4. #3
    Points: 119, Level: 2
    Level completed: 38%, Points required for next Level: 31

    Location
    Glasgow
    Posts
    4
    Thanks
    2
    Thanked 0 Times in 0 Posts

    Re: (SPSS) External validation for a linear regression model

    Hello hlsmith, thank you for your response.
    I checked the linear assumption with residual plots from bivariate regression.
    Both samples are ad-hoc samplings from 2 different cities, as the measurement couldn't be taken place without occupants' permission and they are random.

    Do you mind clarifying which information from descriptive stats for comparison please?

  5. #4
    Omega Contributor
    Points: 38,303, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,993
    Thanks
    397
    Thanked 1,185 Times in 1,146 Posts

    Re: (SPSS) External validation for a linear regression model

    Comparing the covariates that will be introduced into the model and any others ot interest (e.g., means and percentages, etc.).

    Understanding what is different about the samples.
    Stop cowardice, ban guns!

  6. The Following User Says Thank You to hlsmith For This Useful Post:

    wid (01-03-2015)

  7. #5
    Points: 119, Level: 2
    Level completed: 38%, Points required for next Level: 31

    Location
    Glasgow
    Posts
    4
    Thanks
    2
    Thanked 0 Times in 0 Posts

    Re: (SPSS) External validation for a linear regression model

    Below is the descriptive information on data to build model (63 case studies)

    Statistics
    Window Door Pollutant Humidity
    N Valid 64 64 63 63
    Missing 0 0 1 1
    Mean 229.1149 7152.3175 1344.92 49.84
    Std. Error of Mean 62.22448 1226.98730 74.156 .875
    Median 2.6760 139.9100 1143.00 50.00
    Std. Deviation 497.79582 9815.89839 588.595 6.943
    Variance 247800.676 96351861.201 346444.332 48.200
    Skewness 2.075 1.144 .770 .034
    Std. Error of Skewness .299 .299 .302 .302
    Kurtosis 3.405 .778 -.375 -.046
    Std. Error of Kurtosis .590 .590 .595 .595
    Range 1958.04 37732.80 2170 32
    Minimum 1.96 48.00 465 33
    Maximum 1960.00 37780.80 2635 65

    The following is the descriptive information for validation (39 case studies)

    Statistics
    Window Door Pollutant Humidity
    N Valid 39 39 39 39
    Missing 0 0 0 0
    Mean 157.4999 7799.9713 1350.46 45.33
    Std. Error of Mean 58.63950 1500.63454 94.108 1.313
    Median 3.0300 100.1200 1151.00 45.00
    Std. Deviation 366.20359 9371.45971 587.705 8.199
    Variance 134105.068 87824257.039 345396.676 67.228
    Skewness 1.998 .379 1.405 .118
    Std. Error of Skewness .378 .378 .378 .378
    Kurtosis 2.099 -1.959 1.224 -1.014
    Std. Error of Kurtosis .741 .741 .741 .741
    Range 1037.99 18840.34 2487 29
    Minimum 2.54 50.06 607 31
    Maximum 1040.53 18890.40 3094 60


    I also did interaction on window*door to check ( using training data set 63 case studies). Although there is a correlation but result from linear regression analysis did not give any significance. So I did not include this in the model.

  8. #6
    Omega Contributor
    Points: 38,303, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,993
    Thanks
    397
    Thanked 1,185 Times in 1,146 Posts

    Re: (SPSS) External validation for a linear regression model

    Did anything pop out at you presentation ofdata was a little difficult to read) ?

    Also were variable effects comparable between the models, perhaps they may have less statistical power?
    Stop cowardice, ban guns!

  9. #7
    Points: 119, Level: 2
    Level completed: 38%, Points required for next Level: 31

    Location
    Glasgow
    Posts
    4
    Thanks
    2
    Thanked 0 Times in 0 Posts

    Re: (SPSS) External validation for a linear regression model


    Yesterday, I figured out that samples that aren't used in developing the model (in this case 39 cases) are used to get the final model ignoring p-values and adopted the regression coefficients for the generalization. Is this the right path to arrive at the answer? Hope anybody can clarify this. Many thanks

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats