I have a questions regarding the estimation of a specific interaction through the calculation of differences between predicted probabilities.

My outcome variable (outcome) represents students' transition to post-compulsory education and it may take 3 values: 1 for academic qualifications, 2 for vocational and 3 for dropping out.

In the model below the base outcome is vocational qualifications and I am interested in the outcome "academic qualifications". I interact students' ethnicity (White, Asian, Black) with their expectations (low, medium, high). I control by students' prior grades and sex. Below you have the code and the explanation:

mlogit outcome i.ethnicity##i.expectations grades female [pweight= lweights], baseoutcome(2)

margins, over(ethnicity expectations) at(grades=(1 2 3 4)) predict(outcome(1)) post

I am interested in

margins, over(ethnicity expectations) predict(outcome(1)) post

lincom 1.ethnicity#3.expectations - 1.ethnicity#3.expectations

Then I introduce the option at, to see how the predicted probabilities vary at specific values of grades. E.g.:

margins, over(ethnicity expectations)

The problem arises when I calculate the significance afterwards using lincom, since I do not know how to introduce the "at" option:

lincom 1.ypeth2#3.expw4red - 2.ypeth2#3.expw4red

regressor I1.ypeth2#I3.expw4red not found

So I get an error message, most likely because I did not specify the at option

I do not know how to be able to calculate the significance of the differences in predicted probabilities at specific values of the independent variable "grades".

I will appreciate any help in solving this problem.

Best

Mariña Fernández Reino

Postdoctoral Researcher at Universidad Carlos III de Madrid (Spain)]]>

I have a questions regarding the estimation of a specific interaction through the calculation of differences between predicted probabilities.

My outcome variable (outcome) represents students' transition to post-compulsory education and it may take 3 values: 1 for academic qualifications, 2 for vocational and 3 for dropping out.

In the model below the base outcome is vocational qualifications and I am interested in the outcome "academic qualifications". I interact students' ethnicity (White, Asian, Black) with their expectations (low, medium, high). I control by students' prior grades and sex. Below you have the code and the explanation:

mlogit outcome i.ethnicity##i.expectations grades female [pweight= lweights], baseoutcome(2)

margins, over(ethnicity expectations) at(grades=(1 2 3 4)) predict(outcome(1)) post

I am interested in

margins, over(ethnicity expectations) predict(outcome(1)) post

lincom 1.ethnicity#3.expectations - 1.ethnicity#3.expectations

Then I introduce the option at, to see how the predicted probabilities vary at specific values of grades. E.g.:

margins, over(ethnicity expectations)

The problem arises when I calculate the significance afterwards using lincom, since I do not know how to introduce the "at" option:

lincom 1.ypeth2#3.expw4red - 2.ypeth2#3.expw4red

regressor I1.ypeth2#I3.expw4red not found

So I get an error message, most likely because I did not specify the at option

I do not know how to be able to calculate the significance of the differences in predicted probabilities at specific values of the independent variable "grades".

I will appreciate any help in solving this problem.

Best

Mariña Fernández Reino

Postdoctoral Researcher at Universidad Carlos III de Madrid (Spain)]]>

I have a questions regarding the estimation of a specific interaction through the calculation of differences between predicted probabilities.

My outcome variable (outcome) represents students' transition to post-compulsory education and it may take 3 values: 1 for academic qualifications, 2 for vocational and 3 for dropping out.

In the model below the base outcome is vocational qualifications and I am interested in the outcome "academic qualifications". I interact students' ethnicity (White, Asian, Black) with their expectations (low, medium, high). I control by students' prior grades and sex. Below you have the code and the explanation:

mlogit outcome i.ethnicity##i.expectations grades female [pweight= lweights], baseoutcome(2)

margins, over(ethnicity expectations) at(grades=(1 2 3 4)) predict(outcome(1)) post

I am interested in

margins, over(ethnicity expectations) predict(outcome(1)) post

lincom 1.ethnicity#3.expectations - 1.ethnicity#3.expectations

Then I introduce the option at, to see how the predicted probabilities vary at specific values of grades. E.g.:

margins, over(ethnicity expectations)

The problem arises when I calculate the significance afterwards using lincom, since I do not know how to introduce the "at" option:

lincom 1.ypeth2#3.expw4red - 2.ypeth2#3.expw4red

regressor I1.ypeth2#I3.expw4red not found

So I get an error message, most likely because I did not specify the at option

I do not know how to be able to calculate the significance of the differences in predicted probabilities at specific values of the independent variable "grades".

I will appreciate any help in solving this problem.

Best

Mariña Fernández]]>

I'm working with Stata SE 13.1 and using the

I use the following statement to capture the survey details:

pweight: pweight

VCE: linearized

Poststrata: ps_id

Postweight: pw_id

Single unit: missing

Strata 1: region

SU 1: su1

FPC 1: Nh

Strata 2: <one>

SU 2: <observations>

FPC 2: Nph

The issue I'm facing is that when I run estimation commands, such as a

(running mean on estimation sample)

Survey: Mean estimation

Number of strata = 5 Number of obs = 2005

Number of PSUs = 2005 Population size = 581586

N. of poststrata = 231 Design df = 2000

--------------------------------------------------------------

| Linearized

| Mean Std. Err. [95% Conf. Interval]

-------------+------------------------------------------------

jp13 | 3.71035 . . .

--------------------------------------------------------------

Note: missing standard error because of stratum with single

sampling unit.

I'm provided the message "Note: missing standard error because of stratum with single sampling unit.", but having read other posts on this topic, I don't believe the suppression of my results pertains to singleton PSU's. I'm speculating, rather, it pertains to having only sampled only one unit at the second stage.

svydecribe appears as:

Survey: Describing stage 1 sampling units

#Obs per Unit

----------------------------

Stratum #Units #Obs min mean max

-------- -------- -------- -------- -------- --------

1 603 603 1 1.0 1

2 350 350 1 1.0 1

3 351 351 1 1.0 1

4 351 351 1 1.0 1

5 350 350 1 1.0 1

-------- -------- -------- -------- -------- --------

5 2005 2005 1 1.0 1

Can anyone confirm/deny my suspicion and offer any remedial suggestions or theorectical justifications?

Any comments would be appreciated.

Thanks,

Jeremy]]>

First of all I would like to thank everybody for help I found on this forum previously.

Can you please help me with my event study I am trying to do.

My general idea is to perform short-term analysis of the merger outcomes for both bidders and targets.

I have collected data for companies that were involved into merger activities between 1990 and 2014. Afterwards, I collected data for the market indices in countries, where the above-mentioned companies are registered. I reshaped the data into panel data (long format) and combined it, assigning identifiers for companies (as I have companies that were involved into multiple M&A events) and markets. Furthermore, I calculated returns for both companies and market indices. Summarising, I have three datasets at the moment: dataset with companies' returns, dataset with markets' returns and dataset with merger event dates, market identifiers and company identifiers.

The first two files are organised in the form of panel data in long format (first file - [event_id, bidder_name, return]; second file - [market_id, market_name, return]).

You can see the table with my event information variables below:

obs: 4,301 | |||

vars: 9 | 28 Nov 2014 17:46 | ||

size: 554,829 | |||

storage | display | value | |

variable name type | format | label | |

event_id int | %8.0g | General ID | |

Bidder str64 | %64s | Name | |

company_id float | %9.0g | group(Bidder) | |

market_name str44 | %44s | Acquiror Market | |

market_id float | %9.0g | group(market_name) | |

date_announced float | %td | ||

date_effective float | %td | ||

Sorted by: event_id |

According to my idea, I need to perform several steps:

1) I need to combine all the three files together. Can you please give advice on how it can be better done (taking into account further details)?

2) Afterwards, I need to match the merger event dates with the trading dates and calculate abnormal returns for each day in the event window (say, for period of -30 and +30 days around merger event date). The abnormal return is simply difference between market return and company return. Then, I need to sum up all abnormal returns during the event window and obtain CARs (cumulative abnormal returns) for each company. Can you please help me with building a code for this task?

I will appreciate your help. I am looking forward to hearing from you.

Best regards,

Misha Iasinskyi,

PhD Finance,

Nottingham Business School

]]>

Thanks.]]>

Thanks]]>

I have used the following command for a random slopes model:

:

xtmixed OUT i.GRP##c.CONC || id: CONC, mle variance

1 (red): the intercept of GRP0 at CONC 0

2 (green): the p-value for null hypothesis that there is no difference between GRP0 (the reference) and GRP1. This indicates the null hypothesis should be rejected: there is a difference. But is this just at baseline, or across all concentrations?

3 (blue): I am not sure here. Is it the p-value for the null hypothesis that there is no difference in OUT

4 (purple): p-value for the interaction - highly significant, suggesting there is an interaction and that OUT varies with CONC differently for each group

Thanks

Jem

[ATTACH=CONFIG]temp_1199_1417199114040_150[/ATTACH]

[ATTACH=CONFIG]temp_1201_1417199161433_323[/ATTACH]]]>

does Stata support a function which computes a random effectrs model for a multivariate logit regression?

Thank you!]]>

Thanks

esther]]>

esther]]>

Thank you! Andy]]>

In particular, I have observations on three variables:

ID var1 var2 var3

1 1 2 1

2 2 3 1

3 5 4 5

4 6 6 6

...

And I would like to produce a table showing only different patterns and

their observed frequencies, ie.

var1 var2 var3 frequency

First pattern 1 1 1 54

Second pattern 1 1 2 43

-------------------------------

]]>

The specifics:

I'm looking at a test subjected to patients just before an operation, and one year after an operation. Authors just put the mean preoperative score (plus/minus) the SD and the mean one year score (plus/minus) the SD, and the the p-value.

I'd like to do a comparison of the values across the studies (there are more than 5 studies). I don't have any raw data, apart from the mean and the SD.

Any idea how I do this in STATA?

Obviously I looked around quite a bit, and so far i'm looking at Bonferroni and Holm-Sidak, but are these ok to use? And how do I type the command in STATA without having any data in the data editor?

I really hope you can help me!

Sincerely

Bjørn

Aarhus University Hospital, Denmark

]]>