5
$\begingroup$

I'm trying to run a multiple logistic regression model where the dependent variable is dichotomous and independent variables are either bynary or continuous. At first I only had access to PSPP, but ended up getting access to a SPSS license. I ran the model on both softwares with the same dataset and got very different results, and cannot for the life of me figure out why.

Sample size is 4296 records. I have 24 independent variables. I have checked for colinearity. The dataset is somewhat imbalanced, with 190 entry points classified as 1.

  • On PSPP, pseudo-R2s are negative and close to 0 (which I didn't think was even a possibility) - but the performance (% of correct classifications) is pretty good, 86%
  • On SPSS, pseudo-R2s are better, but the performance is poor, closer to 20% of correct classifications
  • On each solution, differente independent variables are presented as significant, and show different (and sometimes contrary) exponents.

Any ideas on what might be going on? I should mention I'm pretty new to all of this.

SPSS model: https://drive.google.com/file/d/1RahEtsR3CertVIrTb3C3iLWAEis6NnFU/view?usp=drive_link

GET
FILE='/Volumes/GoogleDrive-104327187730824169032/O meu disco/_MESTRADO PP
2022 23/_ PESQUISA_LA/DADOS/SPSS resultados/Modelo_com_cortes.sav'.
DATASET NAME DataSet1 WINDOW=FRONT.
LOGISTIC REGRESSION VARIABLES OODC_pop50_provavel
/METHOD=ENTER Pop_log Servidores_logRede Fibra Intranet CPD Cadastro Cad
astro_inf Cadastro_geo
Cadastro_anualPGV PGV_inf PGV_10anos Proporção_de_quadro_com_ensino_su
perior_ou_maisSoftware
Secretaria_ou_instituto_de_planejamentoZoneamento Obras Contribuição D
ependencia_financeiraIPTU
Consorcio Conselhos Proporção_de_CCs_com_ensino_superior_ou_mais
/CONTRAST (Rede)=Indicator
/CONTRAST (Fibra)=Indicator
/CONTRAST (Intranet)=Indicator
/CONTRAST (CPD)=Indicator
/CONTRAST (Cadastro)=Indicator
/CONTRAST (Cadastro_inf)=Indicator
/CONTRAST (Cadastro_geo)=Indicator
/CONTRAST (Cadastro_anual)=Indicator
/CONTRAST (PGV)=Indicator
/CONTRAST (PGV_inf)=Indicator
/CONTRAST (PGV_10anos)=Indicator
/CONTRAST (Software)=Indicator
/CONTRAST (Secretaria_ou_instituto_de_planejamento)=Indicator
/CONTRAST (Zoneamento)=Indicator
/CONTRAST (Obras)=Indicator
/CONTRAST (Contribuição)=Indicator
/CONTRAST (IPTU)=Indicator
/CONTRAST (Consorcio)=Indicator
/CONTRAST (Conselhos)=Indicator
/PRINT=CI(95)
/CRITERIA=PIN(0.05) POUT(0.10) ITERATE(20) CUT(0.5).
enter code here

PSPP model: https://drive.google.com/file/d/1mZMEQy4_kkaoE29eeDtw3rDJGCV57mkd/view?usp=drive_link

GET FILE="/Users/guilhermeiablonovski/Google Drive/_ PESQUISA_LA/DADOS/SPSS resultados/
Modelo_com_cortes.sav".
LOGISTIC REGRESSION OODC_pop50_provavel WITH Pop_log Servidores_log Rede Fibra Intranet
CPD Cadastro Cadastro_inf Cadastro_geo Cadastro_anual PGV PGV_inf PGV_10anos
Proporção_de_quadro_com_ensino_superior_ou_mais Software
Secretaria_ou_instituto_de_planejamento Zoneamento Obras Contribuição
Dependencia_financeira IPTU Consorcio Conselhos
Proporção_de_CCs_com_ensino_superior_ou_mais
        /CATEGORICAL = Rede Fibra Intranet CPD Cadastro Cadastro_inf Cadastro_geo
Cadastro_anual PGV PGV_inf PGV_10anos Software Secretaria_ou_instituto_de_planejamento
Zoneamento Obras Contribuição IPTU Consorcio Conselhos
        /CRITERIA = CUT(0.5) ITERATE(20)
        /NOORIGIN.
$\endgroup$
9
  • $\begingroup$ What's your sample size? How many IVs have you got? Have you checked for collinearity? $\endgroup$
    – Peter Flom
    Dec 3, 2023 at 20:04
  • $\begingroup$ Sample size is 4296 records. I have 24 independent variables. I have checked for colinearity. $\endgroup$ Dec 3, 2023 at 20:08
  • 2
    $\begingroup$ I don't use either SPSS or PSPP, but check whether SPSS uses a different choice for the reference category of a categorical predictor than PSPP. That could at least account for "contrary" coefficient signs. See this page. $\endgroup$
    – EdM
    Dec 3, 2023 at 20:27
  • $\begingroup$ @EdM made a good suggestion, but I checked that and it seems like they are the same. I'm clueless as to what is going on here. $\endgroup$
    – Peter Flom
    Dec 3, 2023 at 21:00
  • 3
    $\begingroup$ I would try just one predictor, and see if you get the same result. Then add some more. See when it breaks down. $\endgroup$ Dec 3, 2023 at 21:16

2 Answers 2

3
$\begingroup$

You have specified different variables in the two models.

Diff_CC_quadro_normalizadois in the PSPP model but not the SPSS model and Proporção_de_CCs_com_ensino_superior_ou_mais is in the SPSS model but not the PSPP model.

$\endgroup$
1
  • $\begingroup$ Good catch, I had the link for a previous iteration of the PSPP tests. Changed for the correct one without that variable difference. Problem persists though. $\endgroup$ Dec 4, 2023 at 11:21
0
$\begingroup$

I ran the model using the same .sav file on Stata this time, and got results similar to the ones provided by SPSS. Since it wasn't possible to identify what was producing the different results in PSPP, I'll simply ignore them.

$\endgroup$
1
  • $\begingroup$ Only similar or essentially the same? $\endgroup$
    – rolando2
    Dec 12, 2023 at 14:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.